Unlocking the Future of Banking with AI

In the fast-paced world of finance, where digital transformation and regulatory changes are constant, keeping up with the latest advancements is essential for success. The financial landscape is being reshaped by emerging technologies, evolving customer expectations, and increasing competition. Banks and financial institutions must navigate these changes while ensuring compliance, managing risks, and delivering exceptional customer experiences.

In our latest LinkedIn article, part of the ai:sight newsletter, we interview Daragh Morrissey, Director of AI at Microsoft Worldwide Financial Services, who shares his perspective on the transformative power of AI in banking. With 15 years at Microsoft, Daragh has witnessed many changes, but none as impactful as GenAI. He highlights how AI is now more accessible, enabling banks to tackle repetitive tasks, enhance fraud detection, and even develop self-driving financial management tools.

Key benefits of AI in banking:

Enhanced Customer Experience: AI-powered chatbots and virtual assistants are revolutionizing customer service by providing 24/7 support and personalized interactions.

Risk Management: Advanced AI algorithms are improving risk assessment and fraud detection, ensuring safer and more secure banking transactions.

Operational Efficiency: Automation of routine tasks through AI is streamlining operations, reducing costs, and increasing efficiency.

Data-Driven Insights: AI is enabling banks to harness the power of big data, offering deeper insights into customer behavior and market trends.

These advancements are not just theoretical; they are being implemented by leading banks worldwide, showcasing the tangible benefits of AI in the financial sector.

To dive deeper into how AI is reshaping the banking industry and to explore real-world examples, check out our full article on LinkedIn.

From Storage to Insights: Streamlining Your Data Pipeline with Redshift and S3

In today’s data-driven world, businesses are constantly generating vast amounts of information that must be analyzed quickly and effectively. Think about companies like Netflix, which processes user behavior data to recommend movies and shows you are likely to enjoy. Or take ride-hailing services like Uber, which rely on real-time data analysis to optimize routes and match drivers with passengers. In these scenarios, data is not just stored—it’s actively shaping the customer experience, often in real time.

As businesses grow, so does their data, turning the challenge from mere storage into ensuring that data is readily accessible for timely analysis. The larger the datasets, the more crucial it becomes to handle them efficiently and at scale.

This is where Amazon Redshift steps in. Redshift is a powerful data warehousing solution that lets businesses run complex queries and perform analytics on massive datasets. For example, a global retailer might use Redshift to analyze purchasing trends across millions of transactions or predict inventory needs in real time. With Redshift, organizations can dive deep into their data, uncover insights that drive smarter decisions, and deliver better customer experiences.

In this guide, we will walk through the process of moving data from Amazon S3 into Amazon Redshift—equipping your team to fully leverage your data for real-time analytics. Let us get started by setting up the necessary infrastructure and permissions.

Step 1: Set up communication between apps

Before we can begin moving data, we need to ensure that our services can communicate securely and efficiently. In AWS, this is done through an Identity and Access Management (IAM) role, which grants the necessary permissions to interact with other AWS services like S3, Redshift, and EC2. Think of it as setting up a key that unlocks access between different services, ensuring they can work together seamlessly.

Creating an IAM role is the first step in our journey, and it will lay the foundation for everything that follows. This role will allow our Redshift cluster to securely access the data stored in S3 and perform the operations we need.

Note: If you’re unfamiliar with generating programmatic access keys, refer to AWS documentation for a guide.

Step 2: Create a Redshift Cluster Using Boto3

With our permissions in place, it is time to create the heart of our data warehousing solution—the Redshift cluster. This cluster is where all the heavy lifting happens, from storing massive datasets to running complex queries.

We will use Boto3, AWS’s Python SDK, to programmatically create this cluster. This approach gives us flexibility and control, allowing us to automate the creation and management of our data warehouse.

However, like all complex operations, there can be challenges. For example, you might encounter an error about a missing default subnet. Do not worry—we will guide you through troubleshooting this issue so you can get your cluster up and running smoothly.

Troubleshooting:

Before moving on, it’s essential to confirm that the Redshift cluster has been successfully created. By doing this, we ensure that everything is set up correctly, avoiding issues down the line. If you run into any problems, such as a missing subnet, we have included steps to resolve them so you can continue with confidence.

AWS_ACCESS_KEY = 'ACCESS_KEY'
SECRET_KEY = 'SECRET_KEY'redshift = boto3.client('redshift', aws_access_key_id=AWS_ACCESS_KEY, aws_secret_access_key=SECRET_KEY, region_name = 'ap-south-1')cluster = redshift.create_cluster(
DBName = "pokemon", #Required
ClusterIdentifier = 'transfer-data-from-s3', #Required
ClusterType = 'single-node',
NodeType = 'dc2.large', #Required
MasterUsername = 'theguywhoknowsnothing', #Required
MasterUserPassword = 'Iwishiknewsomething18' #Required
)

Note: If you are getting an error that “No default Subnet is present….” Then go to the VPC Management Console (select VPC from Services) and create a default in the Region you provided while creating the Boto3 client API.

Troubleshoot Step 1:

Open VPC Management Console from Services.

VPC management console

Troubleshoot Step 2:

Select “Your VPCs” from the left index. Then select “Create default VPC” from the Actions drop down and your default VPC for the desired region will be created.

Before we move on to the next step, let us validate if our cluster has been successfully created or not:

if cluster['ResponseMetadata']['HTTPStatusCode'] == 200:
print("Cluster is generated")

Create default VPC

Step 3: Set Up an Inbound Connection

Now that we have a Redshift cluster, the next step is to establish a secure communication channel to it. Since the cluster operates within a Virtual Private Cloud (VPC), we need to set up an inbound connection to allow data transfer.

This step is crucial because it ensures that the data we move from S3 to Redshift can flow freely and securely. By configuring this connection using an EC2 instance, we create a bridge that links our data storage with our data processing power.

#Create an EC2 resource
ec2 = boto3.resource(
        "ec2",
        region_name = "ap-south-1",
        aws_access_key_id = AWS_ACCESS_KEY,
        aws_secret_access_key = SECRET_KEY
    )#To extract a specific security group we need the VPC ID
#The VPC id can be fetched from the cluster properties
#Using the cluster identifier extract the cluster properties
cluster = redshift.describe_clusters(
        ClusterIdentifier = "redshift-cluster-1"
    )["Clusters"][0]#Let us fetch the VPC ID from the extracted cluster properties
security_group_id = cluster["VpcSecurityGroups"][0]["VpcSecurityGroupId"]
security_group = ec2.SecurityGroup(id = security_group_id)
security_group#Now add the inbound rules to the security group to create the connectionresponse = security_group.authorize_ingress(
        GroupName = security_group.group_name,
        CidrIp = "0.0.0.0/0",
        IpProtocol = "TCP",
        FromPort = 5439,
        ToPort = 5439
    )

Step 4: Transfer Data from S3 to Redshift

With the connection established, we’re ready to move the data. This is where the magic happens—taking raw data from S3 and loading it into Redshift, where it can be transformed into valuable insights.

We’ll use a Python library called psycopg2 to execute the necessary queries on our Redshift database. But before we dive into the transfer, it is important to define the schema of our data. By understanding the structure of what we are working with, like the name and type of a Pokémon in the example below, we ensure that the data is organized and ready for analysis.

Once the schema is defined and our connection is active, we can finally transfer the data from S3 to the Redshift table we created. It is a critical moment, as this step transforms our data from static storage to actionable information.

We will load a single record, saved as a pipe (|) delimited file. This consists of two values, name and type of the Pokémon.

#Connect to Redshift DB now
import psycopg2username = cluster['MasterUsername']
password = <Password>
cluster_endpoint = cluster['Endpoint']['Address']
port = cluster['Endpoint']['Port']
database_name = cluster['DBName']conn_string = "postgresql://{}:{}@{}:{}/{}".format(
    username,
    password,
    cluster_endpoint,
    port,
    database_name
)

con = psycopg2.connect(conn_string)
con

Once we have connected to our Redshift Database, we can now run queries on it:

Time to finally move the data from S3 to the Redshift table we created.

First, make sure the cluster has permission to read from S3, to do this, we navigate to the ‘Properties’ section on the Cluster page, followed by ‘Associated IAM roles’:

Redshift query editor

Create IAM role

From the “Manage IAM roles” select “Create IAM role”. Select permission to read from any or specified S3 bucket. Open the newly generated IAM role and copy the ARN.

sql_query = """
COPY pokemon_team
FROM
's3://pokemon-battle-team-bucket/pokemon_1.txt' /*S3 URI for the file*/
IAM_ROLE 'arn:aws:iam::125094883919:role/service-role/AmazonRedshift-CommandsAccessRole-20220402T223936' /*Open the S3 read access IAM role and copy ARN*/
delimiter '|'
"""execute_query(sql_query, con)#Verify whether the data has moved to the desired table or not
execute_query("select * from pokemon_team", con, True)

Step 5: Delete the Cluster (Optional)

After successfully loading the data, you might want to clean up resources by deleting the Redshift cluster. This step is optional but recommended if you are done with your analysis or want to avoid unnecessary costs.

However, your exploration does not have to end here. The data you have loaded into Redshift is now at your fingertips—ready to be queried, analyzed, and leveraged to drive business decisions. And remember, with AWS services like Redshift Spectrum and Athena, you do not always need to load data into a database to query it; sometimes, you can analyze it directly from S3.

#Note: you can pass in SkipFinalClusterSnapshot as False, if you want to keep a snapshot of your current cluster safe and this way you can resume from where you left of. You also need to specify FinalClusterSnapshotIdentifier in this case.
redshift.delete_cluster(
ClusterIdentifier = "redshift-cluster-1",
SkipFinalClusterSnapshot = True,
)

Feel free to play along with the data you loaded into your Redshift DB. Also, just FYI we do not always need to load data into a database to query it. Thanks to services like Redshift Spectrum and Athena, we can query right off S3 buckets.

Need Help with Implementation?

If you’d prefer to focus on growing your business while leaving the technical setup to experts, we’re here to help.

Our team specializes in designing and deploying data solutions tailored to your unique needs, allowing you to maximize the power of AWS services like Redshift. Reach out to us today, and let’s unlock the full potential of your data together.

Fractal Achieves the AWS Generative AI Competency

We are excited to share that Fractal has been awarded the AWS Generative AI Competency! This recognition by Amazon Web Services (AWS) underscores our commitment to innovation, expertise, and delivering Gen-AI solutions to our clients.

AWS Gen-AI Competency Partners drive the progress of services and tools required for implementing generative AI technologies, and their unique skill set enables them to create and launch transformative applications across various industries.

Achieving the AWS Generative AI Competency sets Fractal apart as an AWS Partner that has demonstrated technical proficiency and customer success in areas such as minimizing hallucinations, prompt engineering, and model customization. Our extensive experience and expertise are evidenced through successful projects that address customer challenges using generative AI solutions. These solutions are pivotal in driving digital transformation strategies that enhance customer experiences, deliver hyper-personalized and engaging content, streamline workflows, and provide actionable insights powered by AWS’s generative AI technology.

What’s Next?

We will continue to innovate, collaborate with AWS, and deliver AI-powered solutions that drive success for our clients. Stay tuned for more updates on our Gen-AI initiatives, client success stories, and upcoming projects.

Learn More about Fractal’s GenAI-powered Knowledge Assist on AWS Marketplace.

About Fractal

Fractal is an Amazon Web Services Partner member of the Amazon Accelerate program and one of the most prominent providers of Artificial Intelligence to Fortune 500® companies. Our vision is to power every human decision in the enterprise and bring AI, engineering, and design to help the world’s most admired companies.

Fractal can deliver solutions with AWS professional services across geographies thanks to a global AWS MSA. Our competency spans AWS services and industries, including data & analytics and the retail industry.

Fractal’s businesses also include Crux Intelligence (AI-driven business intelligence), Eugenie.ai (AI for sustainability), Asper.ai (AI for revenue growth management), and Senseforth.ai (conversational AI for sales and customer service). Fractal incubated Qure.ai, a leading player in healthcare AI for detecting Tuberculosis and Lung cancer.

A comprehensive guide to adopting Microsoft Fabric within your organization

In today’s data-driven world, information is extremely valuable. However, raw data requires the right tools to be transformed into meaningful insights. Microsoft Fabric offers a refined approach to data management, bringing together integration, engineering, warehousing, and analytics into a unified platform. Imagine having a single source of truth for all your data, empowering you to make data-driven decisions with confidence.

Implementing Microsoft Fabric opens exciting opportunities for your organization. Our latest eBook, “A comprehensive guide to adopting Microsoft Fabric within your organization,” offers a clear roadmap to navigate the complexities and maximize the benefits of Fabric. Discover practical strategies, best practices, and real-world examples to guide your organization’s Fabric journey.

Learn how to:

Overcome common adoption blockers
Optimize your data strategy for maximum impact
Leverage Fabric’s advanced features to drive business growth
Build a strong foundation for a data-driven culture

Ready to unlock the full potential of your data? Download our free eBook today and start transforming your business with Microsoft Fabric.

Getting started with the Databricks Unity Catalog

What is the Databricks Unity Catalog?

The Databricks Unity Catalog is a unified governance solution designed for managing data and AI assets within the Databricks Lakehouse Platform.

It offers a centralized platform that enables organizations to efficiently manage, audit, and secure their data assets across different cloud environments. This unified approach ensures robust data governance and compliance, enhancing the overall integrity and security of data operations.

With Unity Catalog vs. Without Unity Catalog

Why should we use the Databricks Unity Catalog?

The Databricks Unity Catalog offers several compelling reasons for its adoption, making it an essential tool for organizations looking to enhance their data governance and management capabilities. Here are the key benefits:

Centralized metadata management

The Unity Catalog provides a central location to manage metadata for all data assets, including tables, views, and machine learning models. This centralized approach simplifies metadata management, ensuring consistency and ease of access across the organization.

Fine-grained access control

The Unity Catalog allows for detailed access controls at the table, row, and column levels. This granularity ensures that only authorized users can access specific data, enhancing security.

It enables data masking and row-level security for sensitive data, protecting confidential information from unauthorized access.

These features ensure that data is accessed securely and in compliance with organizational requirements, maintaining data integrity and privacy.

Data lineage

The Unity Catalog tracks the data flow from source to destination, providing transparency and traceability for data transformations and usage. This visibility is crucial for understanding data dependencies and ensuring data accuracy.

It assists in impact analysis, debugging, and compliance reporting, making it easier to identify and resolve issues, as well as to meet regulatory requirements.

Integration with BI tools

The Unity Catalog seamlessly integrates with popular BI tools like Power BI and Tableau, facilitating easy data consumption and analysis. This integration enables users to leverage their preferred BI tools while benefiting from the robust data governance provided by the Unity Catalog.

By leveraging the Databricks Unity Catalog, organizations can achieve a higher level of data governance, security, and compliance, ultimately enhancing their data management capabilities and driving better business outcomes.

What is an example of a use case?

Scenario

A financial services company, FinCorp, manages large volumes of sensitive data, including customer information, transaction records, and financial reports. Ensuring data privacy, security, and compliance with regulations like GDPR and CCPA is crucial. FinCorp also needs to streamline data access for analysts, data scientists, and business users while maintaining strict access controls.

FinCorp objectives

Centralize Data Governance: Manage data governance policies across multiple Databricks workspaces.
Enhance Security: Implement fine-grained access controls and data masking for sensitive information.
Ensure Compliance: Track data lineage and audit data access for regulatory compliance.
Improve Data Discovery: Provide a searchable catalog of data assets for users.
Facilitate Collaboration: Enable seamless data sharing and collaboration across teams.

FinCorp’s implementation of Unity Catalog

Step 1: Setup the metastore and attach the workspaces

FinCorp currently operates multiple Databricks workspaces, each with its own Hive Metastore. To streamline this setup, we create a Unity Catalog metastore to centralize metadata storage. Then, we attach all of FinCorp’s relevant Databricks workspaces to the centralized metastore.

Step 2: Define data governance policies

To enhance data organization and security, we create schemas and tables for different data domains, such as customer data and transaction data. We implement fine-grained access controls to restrict access to sensitive data, ensuring that only authorized users can view or modify it. Additionally, we apply data masking to sensitive columns, further protecting confidential information from unauthorized access.

Step 3: Enable data lineage and auditing

To ensure comprehensive data governance, we track data lineage to monitor data transformations and usage. Additionally, we use built-in auditing features to log data access and modifications.

Step 4: Enhance data discovery

To enhance data discovery, we tag and describe data assets to improve their discoverability. Additionally, we utilize the Databricks UI to search and explore data assets.

Step 5: Facilitate collaboration

To facilitate collaboration, we enable data sharing across teams using Unity Catalog’s access controls. This allows data scientists and analysts to collaborate on data projects while ensuring data security.

Benefits

Centralized governance: FinCorp has established a unified platform for managing data governance policies, reducing complexity and ensuring consistency across the organization.
Enhanced security: The implementation of fine-grained access controls and data masking techniques ensures the protection of sensitive information.
Regulatory compliance: Detailed data lineage and comprehensive audit logs enable FinCorp to adhere to regulatory requirements and facilitate efficient audits.
Improved data discovery: Users can effortlessly locate and comprehend available data assets, boosting productivity.
Seamless collaboration: Teams can collaborate efficiently without compromising data security.

Use case conclusion

By implementing the Unity Catalog, FinCorp can effectively manage and govern its data assets, ensuring security, compliance, and efficient data usage across the organization.

In summary

Databrick’s Unity Catalog offers a powerful, centralized solution for data governance in the Databricks Lakehouse Platform. It simplifies the management of data assets, enhances security through fine-grained access controls, and ensures compliance with regulatory requirements.

With features like data lineage, auditing, and a searchable catalog, Unity Catalog empowers organizations to streamline their data governance processes, improve data discovery, and facilitate collaboration across teams. By adopting Unity Catalog, organizations can effectively manage their data assets, ensuring they are used efficiently and securely.

Ready to get started?

Are you ready to take the next steps with Databrick’s Unity Catalog? Fractal can help. Contact us to get started.

Read more about the Fractal – Databricks partnership here.

Fractal’s Trial Run: Elevating Retail Excellence at Microsoft Innovation Hubs

TR at the MTC LI post

We are thrilled to announce a significant milestone in our partnership with Microsoft. Fractal’s retail business experiments solution, Trial Run, is now featured as a self-standing demo at all Microsoft Innovation Hubs globally. We are proud to stand alongside Microsoft’s and other leading partners’ technologies, showcased in over 40 Hubs across the Americas, Europe, Asia, and Australia.

Trial Run is an Azure-based SaaS solution for in-store, in-market, and shopper-focused business idea testing. Powered by AI and advanced analytics, it delivers unique, actionable, and statistically significant insights that can empower retailers’ various business functions.

Trial Run enables retailers to build, develop, and scale their experimentation capabilities efficiently and affordably. It also allows businesses to make informed, data-backed decisions on pricing, promotions, and merchandising.

We invite retailers to experience firsthand Trial Run capabilities under the guidance of Microsoft’s Technology Architects during their next Hub visit.

Together, let’s shape the future of retail, one experiment at a time.

Learn more about Trial Run on Azure Marketplace.

Azure Notification Hubs: Key features and use cases

In the digital ecosystem, the efficiency of communication directly correlates with user engagement and satisfaction. Azure Notification Hubs serve as a pivotal element in this communication chain, offering developers a scalable and reliable platform to send notifications across various devices and platforms.

Azure Notification Hubs is a highly scalable mobile push notification engine by Microsoft that is compatible with all significant platforms like

APNs (Apple Push Notification Service)
GCM (Google Cloud Messaging)
WNS (Windows Push Notification Service)
MPNS (Microsoft Push Notification Service)

The Notification Hub can filter push notifications based on a customer, location, or any custom-defined tag.

This capability enables quick delivery of millions of notifications to iOS, Android, Windows, or Kindle devices.

With the Notification Hubs tagging functionality, you can send push alerts to everyone at once or narrow it to specific customer devices. Tags allow you to segment users by activity, interest, geography, or preference, ensuring that the right content gets delivered to the correct person at the right time. A tag could be up to 120 characters long string that includes alphanumeric and non-alphanumeric characters.

Azure Notification Hubs: Key features and use cases The notification tagged with Data_Scientist only reaches the device registered with the tag Data_Scientist, as shown in the diagram.

While sending a push notification with Notification Hubs, tag expressions allow you to target specified groups of devices or, more precisely, registrants. We can target individual device registrations only by associating tags with specific device registrations and then targeting those tags.

Both commercial and consumer scenarios benefit from Notification Hubs. Here are a few things you can do:

Send minimal latency breaking news alerts to millions of people.
Send coupons to user segments who are interested based on their location.
For media, sports, social, gaming, or e-commerce apps, send event or deals-related notifications to users or broadcast to groups.
To attract and advertise to customers, push promotional information to apps.
To notify a user about the status of an activity performed.

Push notifications play a crucial role in boosting user engagement in consumer apps and delivering real-time business updates in enterprise applications. They are efficient on mobile devices, customizable for senders, and remain accessible even when apps are not active, making them the preferred communication service from apps to users.

Below in this article, we will cover:

Configuring the Notification Hub for APNs
Prerequisite for Registering / Viewing / Deleting the devices with Notification Hub,
Registering the devices,
Viewing the registered devices,
Deleting the registered devices.

1. Configuring the Notification Hub for APNs

Prerequisites:

A Notification Hub deployed in the Azure portal. In case you need to create a Notification Hub refer to this Documentation.
In case of Certificate-based authentication, you must have a .p12 certificate file available with the password.
In the case of Token-based authentication, you will require Team ID, Bundle ID, Key ID, and a Token value from a .p8 certificate file.
You can generate the above prerequisite details for authentication modes by following this Documentation.

To configure the Notification Hub with Apple Push Notification Services, we have two Authentication modes available as below:

Certificate-based
Token-based

You can select either Certificate or Token,

For Certificate-based mode:

Select the file icon and select the .p12 file from your local system that you want to upload.
Enter a password associated with the certificate file.
Select Sandbox mode while in development, select Production mode to send push notifications to users who purchased your app from the store.

Azure Notification Hubs: Key features and use cases

For Token-based mode:

Enter the values for Key ID, Bundle ID, Team ID, and Token (obtained from a .p8 certificate file).
Like Certificate-based mode, Select Sandbox mode while in development and select Production mode to send push notifications to users who purchased your app from the store.

Azure Notification Hubs: Key features and use cases

2. Prerequisite for Registering / Viewing / Deleting the devices with Notification Hub

To Register, View, or Delete the devices with the Notification Hub, we have HTTP’s PUT, GET and DELETE REST API methods. The first and foremost requirement to access these APIs is to generate the SAS Token, used as Authorization Header.

The Authorization Header should consist of Token generated as specified in SAS Authentication and authorization with Microsoft Azure Active Directory Access Control.

Microsoft provides a guide to generate SAS tokens using multiple programming languages for using Azure Service Bus REST APIs in this Document.

For the Python approach to generate SAS tokens for using Azure Service Bus REST APIs, we would require a few details as mentioned below:

Namespace Name: The Namespace resource name associated with Notification Hub (namespace could contain one or multiple hubs).
Notification Hub Name: The Notification Hub resource name.
SAS Name: The Shared Access Key Name.
SAS Value: The Shared Access Key Value.

You can find the Namespace name and Notification Hub name from the portal when you navigate to Notification Hub in the overview section, as shown below:

Azure Notification Hubs: Key features and use cases

You can find those details for the SAS name and value parameters when navigating to Access policies under the Manage section in the left panel.

You can find two policies with different permissions. We require the details from the “DefaultFullSharedAccessSignature” policy since it has Full permissions (Listen, Manage, Send).

Azure Notification Hubs: Key features and use cases

We require copying the connection string from DefaultFullSharedAccessSignature policy, which looks like something shown below and contains the Shared Access Key name and its value.

Azure Notification Hubs: Key features and use cases

These details would be required to generate the SAS Token as Authorization Header for accessing APIs.

3. Registering the devices with Notification Hub

Using a Registration or Installation approach, we can accomplish the registration of devices with a Notification Hub.

An installation/registration will associate the PNS handle with tags that we maintain to send a notification to the appropriate user rather than broadcasting them. In the case of APNs, the PNS handle is a unique device token for each device, and to segregate device handles; we can then use Tags.

Note: Notification Hubs allow a maximum of 60 tags for each device.

Compared to the Registration approach, an Installation approach is a much-enhanced approach to registering a device that provides multiple push-related properties. It is the recently developed and enhanced approach to registering your devices with Notification Hub.

Microsoft provides REST APIs for registration and installation, which are user-friendly and require minimal details. Just a Device Token for APNs is enough to register the device with Notification Hub. At the same time, the Tags are an optional element.

Below are a few critical points discussed to choose Installation over Registration for registering the device with Notification Hub:

We could use installation REST APIs to repeatedly make that same call while producing the same result, without any second thought of creating Duplicate registration. It would overwrite the existing registration while reflecting the latest change, if made any.
The Installation approach provides an exceptional Tag format $InstallationId:{<INSTALLATION_ID_GOES_HERE>}, which enables the Notification Hub to classify the unique devices and send the notifications to particular devices rather than broadcasting them.

Suppose a user provides Installation id as testinstallation123 for a particular device while registering with Notification Hub. He could then direct the message to the specific device only by applying $InstallationId:{testinstallation123} Tag, without even doing any extra programming stuff to do so.

This article would move forward with the Installation Approach and a popular tool like Postman to make API calls to register, view, and delete the devices with Notification Hub.

Microsoft provides multiple REST API Methods to perform these tasks, and we would be using Installation REST API for registering the devices with our Notification Hub:

Prerequisites:

Namespace Name
Notification Hub Name
SAS Token for Authorization Header
PNS Handle from iOS device

As discussed earlier in Prerequisite for Registering / Viewing / Deleting the devices with Notification Hub section, we will require the SAS Token generated for Authorization Header, which will look similar to as shown below:

Azure Notification Hubs: Key features and use cases

And the request body would contain parameters and their value as below:

installationid: Any uniquely identifiable string value.
tags(optional): Any string value that can uniquely identify or group the devices (#,:, _ and – are only accepted characters).
platform: APNS.
pushchannel: The PNS handle or device token from the device (device token must be a hexadecimal digit).

The Request Body would look something similar to this:

Azure Notification Hubs: Key features and use cases

Once we send the request, the generated response with status code is 200 with an empty body, which means we have successfully created or overwritten and installation.

4. Viewing the registered devices with Notification Hub

There are multiple possible ways to retrieve the registrations, and you can retrieve all registrations with a Notification Hub or even retrieve selected registrations using the installationid.

MSFT provides multiple REST APIs to deal according to the scenario if we have installationid to retrieve particular registration using Read installation or even retrieve all registrations made with the Notification Hub using Read registrations.

We would retrieve the earlier created registration where we used installationid as 21-2121. To verify if the update gets reflected, we will update its tags and then re-retrieve the registration.

We would make a GET call by providing only the installationid. The only prerequisites here would be the installationid and the SAS Token for Authorization:

Azure Notification Hubs: Key features and use cases

We would now add a new tag to the same registration and see if it reflects the update. To add the tag, we would be using the earlier API that we used to register the device, as it will then overwrite an installation.

Azure Notification Hubs: Key features and use cases

To verify the latest change reflects, let’s now retrieve the installation.

Azure Notification Hubs: Key features and use cases

We were able to retrieve the registered device using the installationid with the latest reflected change.

We would use the similar API below that would return all registered devices with Notification Hub:

Azure Notification Hubs: Key features and use cases

We can even see that the installation has generated an exceptional Tag format, which we discussed earlier $InstallationId:{21-2121}, which would help Notification Hub to classify the unique devices and send the notifications to particular devices rather than broadcasting them.

5. Deleting a registered device

To delete multiple registrations in a single go, we currently don’t have any REST API. Possibly, there are two ways to delete a registered device using either the installationid or registration id.

We would demonstrate by attempting to delete a registered device using the installationid as 21-2121, using the DELETE Method as discussed in this Documentation.

Azure Notification Hubs: Key features and use cases

The installation was successfully removed, and we could see the DELETE Method returned the 204 status code. Even if the installation was not present / not found, the service will return the status code 204.

Later, when we try to retrieve the same registered device after it gets deleted, the GET method used to retrieve devices would return the status code as 404, stating it did not retrieve the installation as below:

Azure Notification Hubs: Key features and use cases

Now that we have completed this demonstration, you can configure your Notification Hub to send alerts and notify your customers about the deals and offers via a quick, simple notification sent using Azure.

In one of our projects for a client, we leveraged Notification Hubs to notify the mobile device users whenever an image successfully uploads to Azure Data Lake Storage (ADLS) containers. Based on specific parameters set to validate the face in the image, a Machine Learning model would accept/reject the uploaded image and notify the device using Azure Notification Hubs.

So why wait? Empower your app with Azure Notification Hub’s powerful capabilities. Contact us to seamlessly integrate and leverage this solution for targeted customer engagement and real-time business updates.

Enhance user interaction across iOS, Android, Windows, and Kindle devices with Fractal’s expertise in deploying Azure Notification Hub to drive app-to-user communication like never before.

Tenant-to-tenant Power BI data migration

In today’s dynamic landscape, Power BI stands out as a pivotal and indispensable tool. It has swiftly become the go-to choice for organizations seeking robust reporting and comprehensive dashboards. Within its data-rich framework lie the essential insights that drive informed business decisions.

Moreover, there are many scenarios where an organization may need to migrate its Power BI tenant. For example, if Company A acquires Company B or Company B merges with Company A, it is essential to transfer all the Power BI data of Company B to Company A’s tenant. Power BI migration can be tricky as there are limited support options for Power BI migration compared to other tools such as Power Apps or Power Automate.

This is where Fractal comes into the picture with its in-house solution that companies can leverage to migrate their Power BI data from one tenant to another.

What is a tenant?

A tenant represents an organization. It’s a dedicated instance of Azure Active Directory AD that an organization or app developer receives at the beginning of a relationship with Microsoft. For example, this relationship could start with signing up for Azure, Microsoft Intune, or Microsoft 365.

Migrating using a third-party migration tool

Image source: Microsoft

Common challenges of tenant-to-tenant migrations

Tenant-to-tenant migration involves multiple parties and teams within each organization, and these groups will be interdependent on the migration tasks. A roadmap that outlines the task owners and their dependencies must be followed.

When Power BI assets, such as reports, datasets, dataflows, etc., are published in a new tenant, their GUIDs (globally unique IDs) change. Thus, these GUIDs need to be carefully modified to make the dependent assets work.

In addition, when company mergers take place, it is often that their SharePoint sites and SharePoint lists also get migrated. Power BI reports using these SharePoint sites would not work as migration of these SharePoint sites would change the Object IDs on the SharePoint side. So, new Object IDs must be placed on the Power BI side to make it work.

There are many such corner cases in tenant-to-tenant Power BI data migrations. For brevity, we will not delve deeply into them. It is essential that these migrations are handled with careful attention and coordinated effort among stakeholders to ensure their success.

How to perform a tenant-to-tenant Power BI data migration

Pre-requisites

Preferably a global admin in source tenant
Power BI Pro or Premium Per User Licenses as per requirement in both source and target tenant
Apart from personal accounts in both tenants, a common service account to setup refreshes in the target tenant

Instructions

Step 1: Analyses of Power BI data in source tenant

All the information related to the Power BI assets is extracted first. This includes information on workspaces, reports, gateways, workbooks, datasets, etc. This would provide us with the intuition on what all thing needs to be migrated.

Step 2: Download all the reports

Download all the reports via automated scripts and keep them on local system. The reports having incremental refresh implemented can’t be downloaded. For this, we need to ask developers to provide us with the source .pbix files. Make sure that all developers stop their developments before you start downloading the reports to avoid version mismatch.

Step 3: Setting up target tenant

Ask your clients to setup all the infrastructure such as gateways, data sources, and VM in target tenant as they have it in their source tenant. Moreover, they also need to provide access on these gateways and data sources to the person who is migrating.

Step 4: Workspace creation

Create all the required workspaces in target tenant with the same configuration as source tenant.

Step 5: Modify reports

We need to modify reports where source applications are also being migrated such as SharePoint, etc.

For datasets, which are ingesting data from Power BI Dataflows or any other Power BI services asset, we need to modify their asset IDs in reports.

Step 6: Publish reports

Upload/Publish all the reports and datasets in their respective workspaces via automated scripts.

Step 7: Attach the data sources to the datasets

Using the common service account, attach the respective data sources to their datasets using automated scripts. Using the same common account, setup the scheduled refreshes using the automated scripts.

Step 8: Create all related Power BI assets

Create all dashboards, apps, workbooks, scorecards, and others which is created in only Power BI services from scratch manually.

Step 9: Provide access

Provide same users access on workspaces and reports as they had in their source tenants.

Step 10: Power BI capacities

If applicable, attach the workspaces to their capacities in target tenant. But first new capacities need to be setup in target tenant.

Step 11: User Acceptance Testing (UAT)

Ask users to perform UAT and inform in case of any discrepancies.

Limitations

Assets that are created in Power BI services cannot be migrated. They need to be created from scratch in target tenant.

Source tenant capacities cannot be reused in target tenant. New capacities need to be bought and setup.

Why Fractal?

There are a few advantages of working with a Microsoft Solutions Partner like Fractal to help you with your tenant-to-tenant Power BI data migration.

Microsoft does not provide services for Power BI data migration from one tenant ID to another tenant ID. Fractal has the right tools and expertise to support clients looking for such requirements.
There are also many dependencies in the whole process. Fractal has the proper roadmap on the execution of whole migration which includes steps for all the parties and has helped many clients in migrating their data.

Fractal demonstrated its expertise by assisting a client during a company acquisition. The client needed to migrate over 1500 Power BI reports to the parent tenant. Fractal not only devised a tailored execution plan for the migration but also completed it ahead of schedule. Remarkably, after the migration, end users continued to enjoy a seamless experience with the reports in the parent tenant, just as they did in the old tenant.

If you want to learn more about tenant-to-tenant Power BI migration? Contact us to speak to one of our experts.

References:

Governing Large Language Models through enhancing techniques

Governing LLMs through enhancing techniques

Generative AI has been transformative across industries, helping businesses to accomplish tasks quicker and more efficiently than ever before. It has been used to create text and visual content, get answers to questions, generate code, and more. Though generic large language models (LLMs), such as the models available through Amazon Bedrock, are incredibly robust and valuable for accomplishing a variety of tasks, there are techniques that organizations can use to improve the quality of their computer-generated content.

Starting with the end-user, the first method of improving the output of generative AI is to craft the perfect prompt through so-called prompt engineering. Organizations can also help their users get more precise output by governing the LLMs themselves. There are two ways an organization could do this by combining approaches across two axes:

Context optimization: Providing more specific and detailed information for the LLM to draw from.
LLM optimization: Adapting the LLM to accomplish specific tasks or work within a given domain.

Let’s go into more detail about how organizations can maximize the effectiveness of their generative AI applications below.

Governing LLMs through enhancing techniques

Prompt engineering

Prompt engineering refers to crafting specialized and effective prompts to guide language models and optimizing the context for the desired outputs. Put simply, that means writing the prompt in a way to get exactly what is expected. Here are five guidelines to help end users with their prompt writing.

Be specific and contextualize: Clearly define the topic and provide context. Specify the audience, tone, and timeframe. The more precise the prompt, the better the AI’s response will be.
Iterate for refinement: Use an iterative approach. Refine the prompt based on initial outputs. Adjust and experiment to get more useful information.
Include an action or task: Add an action or task for the AI to complete. For instance, ask it to summarize, outline, list, plan, compare, or predict. This helps guide the AI toward a specific outcome.
Set parameters: Specify parameters such as word count, exclusions, or formatting preferences (e.g., subheadings, paragraphs). This ensures the AI generates content that aligns with the end-user’s requirements.
Avoid bias: Be mindful of bias. Biased prompts can lead to biased outputs. For example, asking a leading question may produce a skewed answer. Aim for balanced prompts that explore both sides of an issue.

From an organizational perspective, companies can help end-users maximize the use of their GenAI apps by providing instruction and guidance on how to properly craft prompts for a given task. Generative AI is a new technology that many end-users are still getting used to. Giving them resources to learn more quickly is one of the best ways to help them get the most out of the app.

Context Optimization: Retrieval Augmented Generation (RAG)

To further increase the quality and usefulness of AI-generated content, organizations can also provide LLMs with additional information before the prompt so that the LLM executes prompts in a certain way. Called Retrieval Augmented Generation (RAG), this method allows organizations to pre-define the scope or context of the prompt. Here is an example of how it might be used in real life.

Scenario:

Imagine you are an executive at an electronics company that sells devices like smartphones and laptops. Your company wants to provide detailed answers about its products to customers, journalists, and investors. You have a vast knowledge base containing information about various product specifications, features, and use cases.

Customer Query: “What makes our latest smartphone stand out?”

Without RAG:

In this scenario, without RAG the LLM can produce a result on the prompt, however, the LLM lacks specific details about your products beyond what it learned during training.

Response (without RAG): “Our latest smartphone is a great device. It has a powerful processor, a nice display, and a good camera. The battery life is decent too.”

Using RAG:

By incorporating the knowledge base into the LLM’s prior knowledge, when you ask the LLM a question about a specific product, it can retrieve relevant information from the knowledge base and integrate it into the generated response. The result is a much more targeted and specific response.

RAG-Enhanced response: “Our latest smartphone boasts a powerful Snapdragon processor, a stunning AMOLED display, and a triple-camera setup for exceptional photography. Additionally, its long-lasting battery ensures uninterrupted usage throughout the day.”

RAG can be enabled through various approaches, and Amazon Bedrock offers multiple tools that can be used depending on the selected approach.

LLM optimization: Fine-Tuning

To further improve the results of generative AI responses, organizations can fine-tune pre-trained LLM models to specific tasks and domains. There are three ways they can approach this customization:

Supervised Learning Fine-Tuning: 

In supervised learning, organizations provide a predefined correct answer (ground truth) for a specific task.
The LLM is then fine-tuned using labeled data to generate more accurate and contextually relevant responses.
For example, if an organization wants to create an LLM for code completion, they can fine-tune it using a dataset of code snippets and their corresponding completions.

Reinforcement Learning from Human Feedback (RLHF): 

RLHF involves training the LLM based on feedback from human evaluators.
Human evaluators rank different model responses based on quality, relevance, and correctness.
The LLM learns from this feedback to improve its performance over iterations.
RLHF is particularly useful when there is no predefined correct answer, as it allows the model to learn from human preferences and adapt accordingly.

Domain-Specific Fine-Tuning: 

Organizations can fine-tune LLMs on domain-specific data.
For example, a healthcare organization can LLM fine-tuned on medical texts to provide accurate diagnoses or treatment recommendations.
Similarly, in finance or technology, fine-tuning on relevant data enhances the LLM’s utility for specific tasks.

This fine-tuning strikes a balance between generalization and specialization, allowing LLMs to excel in specific contexts while retaining their overall capabilities.

Organizations can fine-tune LLMs for their users by using Amazon Bedrock.

Putting it all together

Of course, these techniques aren’t mutually exclusive. You could use one or both of them depending on your needs and the resources available in your organization.

If you are looking for help in putting the pieces together, Fractal experts can help you plan and implement a solution that will work for you.

Advancing Streaming Piracy Detection and Prevention: A Comprehensive Approach

In the dynamic world of Media & Entertainment, the fight against content piracy is a constant battle. Despite the industry’s efforts to implement legal and technological safeguards, piracy continues to be a significant issue, affecting revenue and exposing consumers to potential risks such as malware and identity theft. The rise in digital piracy is fueled by factors such as the demand for immediate content access, economic challenges, and the uncontrolled use of generative AI technology.

A comprehensive strategy is essential to combat piracy effectively. This strategy should include educating consumers, deploying advanced technology, and establishing strong legal frameworks. It is also vital for industry stakeholders to work together to develop efficient anti-piracy measures and explore new business models that improve content accessibility and user experience.

A cutting-edge anti-piracy solution utilizes AI, scalable data, and cloud engineering to provide real-time insights and protect content. Such a solution, hosted on Microsoft Azure, will need to feature capabilities such as near real-time IP and device monitoring, automated risk scoring, and investigation of suspicious devices to identify and address piracy threats.

Fractal's Anti-piracy framework

To achieve this, a solution needs to be based on three fundamental principles: capturing dynamic user behavior, using AI and ML for anomaly detection and network analysis, and applying design thinking to create user-centric solutions. The outcome will then be a robust, transparent, and actionable anti-piracy system that empowers analysts and streamlines the decision-making process.

By implementing this intelligent anti-piracy framework, a content provider can understand pirate behavior, adapt to changing piracy patterns, and reduce reliance on manual decision-making.

The Media & Entertainment industry must address piracy with a holistic approach. It is crucial for the Media & Entertainment industry to protect content creators and consumers from the negative impacts of content piracy and promote a sustainable digital environment.

Download our eBook to learn more.

Governing Large Language Models through enhancing techniques

Generative AI has been transformative across industries, helping businesses to accomplish tasks quicker and more efficiently than ever before. It has been used to create text and visual content, get answers to questions, generate code, and more. Though generic large language models (LLMs), such as the models available through Google Cloud Vertex AI, are incredibly robust and valuable for accomplishing a variety of tasks, there are techniques that organizations can use to improve the quality of their computer-generated content.

Context optimization: Providing more specific and detailed information for the LLM to draw from.
LLM optimization: Adapting the LLM to accomplish specific tasks or work within a given domain.

Let’s go into more detail about how organizations can maximize the effectiveness of their generative AI applications below.

Governing LLMs through enhancing techniques

Prompt engineering

Be specific and contextualize: Clearly define the topic and provide context. Specify the audience, tone, and timeframe. The more precise the prompt, the better the AI’s response will be.
Iterate for refinement: Use an iterative approach. Refine the prompt based on initial outputs. Adjust and experiment to get more useful information.
Include an action or task: Add an action or task for the AI to complete. For instance, ask it to summarize, outline, list, plan, compare, or predict. This helps guide the AI toward a specific outcome.
Set parameters: Specify parameters such as word count, exclusions, or formatting preferences (e.g., subheadings, paragraphs). This ensures the AI generates content that aligns with the end-user’s requirements.
Avoid bias: Be mindful of bias. Biased prompts can lead to biased outputs. For example, asking a leading question may produce a skewed answer. Aim for balanced prompts that explore both sides of an issue.

Prompt engineering is a practice that is accessible to end users in any generative AI product they use, whether those products use general LLMs or those managed by their organization via Vertex AI.

Context Optimization: Retrieval Augmented Generation (RAG)

Scenario:

Customer Query: “What makes our latest smartphone stand out?”

Without RAG:

In this scenario, without RAG the LLM can produce a result on the prompt, however, the LLM lacks specific details about your products beyond what it learned during training.

Response (without RAG): “Our latest smartphone is a great device. It has a powerful processor, a nice display, and a good camera. The battery life is decent too.”

Using RAG:

RAG-Enhanced response: “Our latest smartphone boasts a powerful Snapdragon processor, a stunning AMOLED display, and a triple-camera setup for exceptional photography. Additionally, its long-lasting battery ensures uninterrupted usage throughout the day.”

RAG can be enabled through various approaches, and Google’s Vertex AI offers multiple tools that can be used depending on the selected approach.

LLM optimization: Fine-Tuning

To further improve the results of generative AI responses, organizations can fine-tune pre-trained LLM models to specific tasks and domains. There are three ways they can approach this customization:

Supervised Learning Fine-Tuning:

In supervised learning, organizations provide a predefined correct answer (ground truth) for a specific task.
The LLM is then fine-tuned using labeled data to generate more accurate and contextually relevant responses.
For example, if an organization wants to create an LLM for code completion, they can fine-tune it using a dataset of code snippets and their corresponding completions.

Reinforcement Learning from Human Feedback (RLHF):

RLHF involves training the LLM based on feedback from human evaluators.
Human evaluators rank different model responses based on quality, relevance, and correctness.
The LLM learns from this feedback to improve its performance over iterations.
RLHF is particularly useful when there is no predefined correct answer, as it allows the model to learn from human preferences and adapt accordingly.

Domain-Specific Fine-Tuning:

Organizations can fine-tune LLMs on domain-specific data.
For example, a healthcare organization can LLM fine-tuned on medical texts to provide accurate diagnoses or treatment recommendations.
Similarly, in finance or technology, fine-tuning on relevant data enhances the LLM’s utility for specific tasks.

This fine-tuning strikes a balance between generalization and specialization, allowing LLMs to excel in specific contexts while retaining their overall capabilities.

Organizations can fine-tune LLMs for their users by using Vertex AI.

Putting it all together

Of course, these techniques aren’t mutually exclusive. You could use one or both of them depending on your needs and the resources available in your organization.

If you are looking for help in putting the pieces together, Fractal experts can help you plan and implement a solution that will work for you.

Fractal’s PYO Autonomous AI solution presentation on Microsoft booth at Hannover Messe ‘24

Fractal is proud to announce that its Production Yield Optimization (PYO) solution will be presented Wednesday, April 24th at 10 am on Microsoft’s stand (Hall 17, Stand G06) at the 2024 Hannover Messe.

PYO is a proven Autonomous AI solution that enables manufacturers to reduce waste and optimize manufacturing output with AI agents that dynamically fine-tune existing control systems parameters.

PYO is built on Azure Machine Learning and leverages industry-standard deep reinforcement learning (DRL) techniques and libraries to translate subject matter experts’ expertise into so-called “reward functions” that allow the AI agent to self-train through a custom-built AI simulator. This expertise transfer is commonly referred to as “Machine Teaching”.

Fractal’s end-to-end DRL experience, accelerators, and best practices help manufacturers customize PYO to their unique needs. This is one of the reasons why companies such as PepsiCo have trusted Fractal for their PYO project on their Cheetos product line.

How does PYO work?

The PYO AI agent design, training, and deployment is a multi-step and iterative process.

PYO technical architecture

Once the initial Machine Teaching process is completed, the manufacturing SMEs expertise will help select the appropriate process training data and define the best reinforcement learning reward function. The AI agent is then trained using the new simulation.

The trained agent is then validated virtually with the simulation. However, in most cases, multiple cycles of simulation fine-tuning, reward function adaptation, agent training, and virtual validation will be required to converge to a satisfactory PYO AI agent. The agent is then trialed and fully deployed in production.

The AI agent sends control signals to the simulation or the production system. The reward function will measure the difference between the expected simulation or system state versus its actual state and will modify the agent’s deep neural network weights accordingly. Depending on the system controlled, this training loop will run between hundreds of thousands to millions of times.

Why should you consider PYO?

Optimize production with AI: PYO agents learn, through SME expertise transfer, to optimize production for complex and changing environments. Those agents will help with both manufacturing line-level and human-level challenges.
Real-world AI solution: PYO uses DRL and simulations to train the AI agents without the need for pre-existing labeled datasets.
Fractal’s end-to-end expertise: Bringing an AI agent from design to deployment requires a large set of data and AI skills. Fractal is a recognized Microsoft Solution Partner with the expertise to support you throughout your data and AI transformation journey.

If you want to learn more about PYO and how it can help you achieve your production goals, come and attend our presentation at the Microsoft booth on Wednesday, April 24^th, at 10 am.

You can also check out our PYO solution page here.

Digital twin vs. simulations: the quick cheat sheet

What is a digital twin?

Wikipedia defines a digital twin as a “virtual representation that serves as the real-time digital counterpart of a physical object or process.”

In theory, a digital twin will gather input from connected sensors, machinery, and people to store and display them in a cloud-hosted application. However, besides being able to look back in time what happened when and perform some post-mortem analysis, a digital twin limited to a backward-looking view won’t have much business interest.

Therefore, often digital twins will also integrate simulations of what is represents. Those simulations can be at the device, process, or even plant level. It will allow users to leverage this combination of real-time data and system-level behavior modeling through the simulator for multiple use cases.

Digital twin vs. simulations: the quick cheat sheet

For instance, a digital twin can be used to: 

Replay a system behavior based on historical data 
Do advanced “what-if” analysis before deciding which path to choose 
Train new operators on virtual processes before letting them work on the actual real-life process 
Simulate a process to train an AI Agent using Deep Reinforcement Learning (DRL)
And more depending on the industry that digital twins are used in

“Digital twins represent a great business improvement opportunity for customers across industries. However, what they are, how they work, how they can positively impact operations, and what technologies are involved with digital twins are often questions customers struggle to answer” – Manish Amin, Data & AI and IoT Principal and Advisor, Microsoft

Digital twin vs. simulator 

A simulator’s scope is often limited to a particular piece of equipment or process, although not always. Once programmed or trained, the simulator will run separately from the real-life process.

Conversely, a digital twin will often encompass a broader process comprised of multiple pieces of equipment, and it will remain connected to the live system to represent it faithfully.

Therefore, a simplified way to think about the difference between a digital twin and a simulator is to consider that a digital twin is a simulation whose states (inputs, outputs) are updated to accurately reflect their real-life value. A simulator could end up drifting from real life or even provide wrong data; a digital twin won’t if it remains connected.

Digital twin vs. simulations: the quick cheat sheet

Conversely, simulators operate separately from a real-life process and can even be developed without an existing process to test hypothesis.

Enabling technologies for digital twins and simulations

To build and run a digital twin, several technology blocks are required.

A simulation engine
Real-time process data collection technologies
Cloud and data services to collect, store and analyze the process and simulation data

How do you build a simulation?

Digital twin vs. simulations: the quick cheat sheet

There are multiple ways to build a simulator, and the three most used are:

Physics-based simulators
Software package-based simulators using products such as AnyLogic or Simulink 
AI or data-based simulators that train AI models, most often deep neural networks-based ones

For the latter approach, data-logging only digital twins can be used to create the dataset necessary to train this AI simulator. The historical process data (both inputs, states, and outputs) that the digital twin recorded can provide the breadth and quantity of labeled data required for those types of AI simulators’ supervised learning. This is one of the areas where a partner with extensive data science experience can significantly help with the speed and quality of the simulation development.

Additional enabling technologies 

Digital twin vs. simulations: the quick cheat sheet

To collect real-time process data, smart sensors using technologies such as Azure IoT are going to be required. Adding intelligence at the edge to existing sensors or deploying new smart sensors such as vision AI ones, we will be able to instrument all the relevant process inputs and outputs.

This real-time data and the simulator(s) will be hosted on an appropriate cloud platform to enable the above-mentioned use cases. Solutions such as Azure Digital Twin will enable easy integration of those elements and access to device, process, line (or building), or plant (or campus) digital twins.

“IoT is inextricably linked with digital twins. To create a comprehensive digital twin of a manufacturing environment, one must connect every major process on a manufacturing floor to IoT for process digitization, modeling, and simulation” – Mohammad Ahmed, Azure Infrastructure Specialist, Microsoft 

Although this article somewhat oversimplifies both what digital twins are and what is required to build and run them, it provides a base for more in-depth research if a more comprehensive understanding of the subject is required. To help with this additional research, we listed a few links below to get you started.  

Also, as Fractal possesses the end-to-end technological capabilities required to instrument, build, train, deploy, and maintain digital twins, feel free to contact us, if you are interested in learning more about this topic. 

Additional resources on digital twins:

Governing Large Language Models through enhancing techniques

Generative AI has been transformative across industries, helping businesses to accomplish tasks quicker and more efficiently than ever before. It has been used to create text and visual content, get answers to questions, generate code, and more. Though generic large language models (LLMs), such as the OpenAI models available through Azure AI, are incredibly robust and valuable for accomplishing a variety of tasks, there are techniques that organizations can use to improve the quality of their computer-generated content.

Context optimization: Providing more specific and detailed information for the LLM to draw from.
LLM optimization: Adapting the LLM to accomplish specific tasks or work within a given domain.

Let’s go into more detail about how organizations can maximize the effectiveness of their generative AI applications below.

Governing LLMs through enhancing techniques

Prompt engineering

Be specific and contextualize: Clearly define the topic and provide context. Specify the audience, tone, and timeframe. The more precise the prompt, the better the AI’s response will be.
Iterate for refinement: Use an iterative approach. Refine the prompt based on initial outputs. Adjust and experiment to get more useful information.
Include an action or task: Add an action or task for the AI to complete. For instance, ask it to summarize, outline, list, plan, compare, or predict. This helps guide the AI toward a specific outcome.
Set parameters: Specify parameters such as word count, exclusions, or formatting preferences (e.g., subheadings, paragraphs). This ensures the AI generates content that aligns with the end-user’s requirements.
Avoid bias: Be mindful of bias. Biased prompts can lead to biased outputs. For example, asking a leading question may produce a skewed answer. Aim for balanced prompts that explore both sides of an issue.

Prompt engineering is a practice that is accessible to end users in any generative AI product they use, whether those products use general LLMs or those managed by their organization via Azure. In the Microsoft technology stack, users can access these LLMs through Copilot which is available as a standalone app and is integrated into Edge and Microsoft 365.

Context Optimization: Retrieval Augmented Generation (RAG)

Scenario:

Customer Query: “What makes our latest smartphone stand out?”

Without RAG:

In this scenario, without RAG the LLM can produce a result on the prompt, however, the LLM lacks specific details about your products beyond what it learned during training.

Response (without RAG): “Our latest smartphone is a great device. It has a powerful processor, a nice display, and a good camera. The battery life is decent too.”

Using RAG:

RAG-Enhanced response: “Our latest smartphone boasts a powerful Snapdragon processor, a stunning AMOLED display, and a triple-camera setup for exceptional photography. Additionally, its long-lasting battery ensures uninterrupted usage throughout the day.”

RAG can be enabled through various approaches, and Azure offers multiple tools that can be used depending on the selected approach: Azure AI Search , Azure AI Studio, Azure OpenAI Studio, and Azure Machine Learning.

LLM optimization: Fine-Tuning

To further improve the results of generative AI responses, organizations can fine-tune pre-trained LLM models to specific tasks and domains. There are three ways they can approach this customization:

Supervised Learning Fine-Tuning:

In supervised learning, organizations provide a predefined correct answer (ground truth) for a specific task.
The LLM is then fine-tuned using labeled data to generate more accurate and contextually relevant responses.
For example, if an organization wants to create an LLM for code completion, they can fine-tune it using a dataset of code snippets and their corresponding completions.

Reinforcement Learning from Human Feedback (RLHF):

RLHF involves training the LLM based on feedback from human evaluators.
Human evaluators rank different model responses based on quality, relevance, and correctness.
The LLM learns from this feedback to improve its performance over iterations.
RLHF is particularly useful when there is no predefined correct answer, as it allows the model to learn from human preferences and adapt accordingly.

Domain-Specific Fine-Tuning:

Organizations can fine-tune LLMs on domain-specific data.
For example, a healthcare organization can LLM fine-tuned on medical texts to provide accurate diagnoses or treatment recommendations.

Similarly, in finance or technology, fine-tuning on relevant data enhances the LLM’s utility for specific tasks.

This fine-tuning strikes a balance between generalization and specialization, allowing LLMs to excel in specific contexts while retaining their overall capabilities.

Organizations can fine-tune LLMs for their users by using Azure AI Search and Azure Machine Learning.

Putting it all together

Of course, these techniques aren’t mutually exclusive. You could use one or both of them depending on your needs and the resources available in your organization.

If you are looking for help in putting the pieces together, Fractal experts can help you plan and implement a solution that will work for you.

References:

Three pillars of the retail industry: Replenishment, allocation, and transportation 

The retail industry is one of the most dynamic and fast-paced sectors, comprised of a variety of stakeholders engaged in selling finished products to end-user consumers. In 2022, the U.S. retail sector was estimated at more than seven trillion dollars. The sector is projected to continue to grow, and by 2026 U.S. retail sales are expected to reach approximately USD 7.9 trillion. With the increased demand for consumer goods in different sectors and the ever-increasing choices of products at low costs, investments in the retail sector have also grown over the past few years.

As there is always an element of change in this industry, new challenges are encountered every day. Take product stockouts, for example. Suppose a customer walks into a grocery store to purchase items of their favorite brand but discovers that the product is not available. Frustrated by this, the customer chooses to either buy another brand or postpone the purchase; both scenarios are unfavorable to the business. The brand image and sales of the product are damaged because of this out-of-stock issue. The out-of-stock situation occurs when the inventory of a particular product is exhausted; this causes a problem for both suppliers and retailers.

There can be multiple reasons that would cause the product stockout, such as inaccurate inventory data, lack of demand forecasting, or an unseasonal spike in purchasing. Many of these underlying causes of stockouts can be avoided if the business implements adequate processes to be carried out every month.

To avoid situations like the stockout example above, retail companies need to develop a methodology for streamlining the following operations:

Replenishment
Allocation
Transportation

3 pillars of retail industry

These three operations create the three pillars of the retail industry that help monitor real-time insight into customer behavior and understand their buying patterns hence strengthening the retail foundation.

Replenishment

Replenishment refers to a situation where the amount of stock left in the store is counted so that the right products are available at an optimal quantity. It is considered an essential aspect of inventory management as it ensures that the right products are being reordered to meet the customer demand.

In operational terms, the efficiency of store replenishment has a significant impact on profitability. The effectiveness and accuracy of store ordering affect sales through shelf availability and storage, handling, and wastage costs in stores and other parts of the supply chain. By optimizing demand forecasting, inventory management, and setting of order cycles and order quantities by making them more systematic, the gains obtained are significant, often amounting to savings of several percent of total turnover.

For companies that must manage a large number of SKUs, one of the most effective ways of making store replenishment more accurate, efficient, and cost-effective is by using a replenishment system specifically tailored to business operations. When many different products need to be managed, manual ordering is highly labor-intensive and expensive; this results in companies using the replenishment system.

An efficient replenishment system reduces process costs, improves inventory turnover, and provides higher service levels. The system constantly monitors the stock, sales, and demand while considering the forecast changes in demand and adjusting the replenishment orders. Recognizing the sales frequency, sales value, or profit margin, the company can control its inventory in such a way that ensures long-term profitability. The replenishment system calculates the safety stock level for each SKU separately and sets them to meet the service level targets with efficiency, considering the predictability of demand.

Allocation

In allocation, the new stock of products is distributed to individual store units, such that they maximize the sale of the product and prevent any stock out situation in the future. This process enables the assigning of supplies so that they support the organization’s strategic goals. Having sufficient stock levels is an essential component for any retail business; with the changing consumer habits, it becomes crucial for the stock to be available in the right place at the right time.

To meet new and increasing demands, retailers need an efficient process to gather, interpret, and analyze data from customer behaviors and habits, which would help get a more localized and specific idea of what is sold at a larger quantity in different locations. Items that are high sellers in one particular area may not sell well in others, so recognizing and monitoring this can ensure that the stock is allocated to the most needed location. Due to this, an opportunity is provided for the retailers to encourage sales by pushing stock of a similar type that a customer may favor at a particular location.

Transportation

Transportation plays a significant role in delivering the right stock of products at the right point of delivery. It connects the business to its supply chain partners and influences the customers’ satisfaction with the organization. With the ever-changing customer preferences and as their expectations continue to evolve, the transportation industry is undergoing a dramatic transformation to meet these demands.

Today, data plays a vital role in shaping how the industry will progress amidst tough competition. Due to the maturation of automation technologies, AI will help the transportation industry to manage drivers and fleet managers. By employing the techniques of AI, fleet and truck adjustments will offer data in real-time, eventually improving the industry’s standard. The safety and retention of the drivers will also increase from these newly acquired standards, and with enhanced access to data, there will be transparent communication between drivers and carriers.

The average time between goods purchasing and delivery decreases by using real-time findings, making retailers focus on transportation to improve their business performance. The ability to automate insights, alerts, and data exchange more quickly will be the game-changer for this industry.

These three pillars of retail can be strengthened by adopting in-house solutions and capabilities like Trial Run, customer analytics, supply chain analytics, and store operation analytics.

How could these solutions help the retail industry?

Trial Run is a data-driven, cloud-based test management product used to test business ideas for sites and markets using Google Cloud capabilities combined with advanced technologies to improve customer experience, enhance product recommendations, streamline operations, optimize inventory, and enhance the supply chain.

Trial Run helps in scientific and systematic testing, which can unlock insights and provide direction with the following tests:

Marketing and Merchandizing tests
In-store experience tests
Store operations tests

Customer Analytics is an AI-driven suite that helps a retailer to know their customers in a better way by understanding the customer needs and preferences from every angle, like acquisition, engagement, retention, and growth, gaining insights that can fuel growth in marketing initiatives, loyalty programs, or eCommerce platforms.

Supply-chain Analytics is an advanced analytics and intelligent automation solution that helps in end-to-end supply chain visibility to stay competitive, keeping the distribution networks customer-oriented and efficient while reducing environmental impact. It helps in streamlined operations, which results in better cost savings, ultimately delivering more value in every step of the supply chain process.

Store Operation Analytics helps boost sales productivity and reduce costs across every facet of store operations – from labor, facilities, and inventory management to enhanced customer service and satisfaction.

All these solutions and capabilities help understand the customer motivations, preferences, and desires to meet their demands and increase sales effectively, hence strengthening the pillars of the retail industry.

Conclusion

To meet these growing customer expectations, retailers should give priority to collecting the customer data and analyzing it to support business decisions throughout their value chain. The inventory stocking patterns and shipping routes will shift in relation to patterns informed by this data. Retailers should make a concentrated effort to leverage the data while making critical business decisions, and to remain efficient; they must remain flexible and transform their operations as they capture more insights from their data.

Over the past 20+ years, Fractal has helped many retail companies make their replenishment, allocation, and transportation operations more efficient by leveraging AI, engineering, and design. If you would like to learn more about optimizing these aspects of your retail business, please contact us to speak with one of our experts.

Find how Trial Run enabled decision-making and helped clients with increased accuracy in test analysis and better ROI measurement resulting in an annual financial gain of $25Mn.

Conduct business experiments on site with Trial Run

Nowadays, companies want to be able to test business decisions and ideas at a scale large enough to believe the results but also at a scale small enough to reduce the large investments and risks that come with full-scale execution.

Trial Run helps conduct tests such as altering store layouts and remodeling, loyalty campaigns, and pricing to recommend the best possible tailored rollout to maximize gains. You can now implement new ideas with minimal risk and maximum insight with the power of business experimentation. Trial run helps you:

Test each business idea at scale to generate customer insights without excessive spending.
Find out why your customers behave the way they do.
Learn how your customers will react to your new big idea.

What is Trial Run?

Trial Run is a data-driven, cloud-based test management platform used to test business ideas for sites, customers, and markets. Trial run is built using Amazon EKS, Amazon Redshift, Amazon EC2, Amazon ElastiCache, and Amazon Beanstalk. It is intuitive for beginners and experts alike and helps companies scale experimentation efficiently and affordably.

Trial Run supports the entire experimentation lifecycle, which includes:

Trail Run illustration

Design: Build a cost-effective and efficient experiment that gives you the data you need to proceed with confidence.
Analyze: Work with variables that provide you with targeted and actionable insights.
Act: Use the generated insights to ensure your new rollout provides your stakeholders with the precise ROI.

Trial Run offers valuable support across various operational and administrative departments, including Retail, Consumer Packaged Goods (CPG), and Telecommunications.

Through its scientific and methodical testing approach, Trial Run can uncover fresh perspectives and guide decision-making through a range of tests, including:

Marketing and merchandising strategies.
Enhancing the in-store experience.
Examining store operations and processes.

These tests are carried out at the store operations and process, product, or consumer levels.

Trial Run offers a dynamic, affordable, and modern way of experimentation so you can stay relevant in a rapidly changing business environment. Trial Run also helps you to drive experiments through:

Driver Analysis: Identify key factors that are significant in driving the business outcomes
Rollout simulator: Maximize the ROI of a campaign
Synthetic Control Algorithm: Determine the right number of control stores with appropriate weights to create the replica of the test store
Experiment calendar: Avoid overlaps in experiments
Clean search: Let Trial Run parse the experiment repository and find entities that are available for a test

What you can expect from Trial Run

Graphical design elements make it easy to use the program as an expert or a beginner
Automated workflows can guide you through the process from start to finish
Highly accurate synthetic control results with automated matching processes that only require minimal human intervention
Experiments at speed and scale without the hassle of expert teams or expensive bespoke solutions
Training, troubleshooting, and best practices from the best in the business
Easy pilots to help your new idea go live in as little as 6 to 8 weeks

Trial Run stands out from other solutions by offering a transparent methodology and easily explainable recommendations. Trial Run utilizes a cutting-edge technique called “synthetic control” for matching, ensuring precise results. Trial Run can be used as a SaaS offering that is easily scalable based on demand and can be hosted on the cloud of customer’s choice. With Trial Run software, customers have unlimited test capabilities, enabling them to design and measure numerous initiatives without any restrictions. Finally, Trial Run success is proven in enterprises, with over 1,000 use cases deployed on our platform.

How do I get started?

Are you ready to implement cutting-edge technology to help you build cost-effective and efficient experiments that provide you with the data you need to make decisions?

If you want to achieve successful Trial Run implementation, get started on the AWS Marketplace.

Interested in learning more about how Fractal can help you implement Trial Run, contact us to get in touch with one of our experts.

Conduct business experiments on site with Trial Run

Test each business idea at scale to generate customer insights without excessive spending.

Find out why your customers behave the way they do.
Learn how your customers will react to your new big idea.

What is Trial Run?

Trial Run is a data-driven, cloud-based test management platform used to test business ideas for sites, customers, and markets. Trial run is built using Azure Kubernetes Services, Azure Synapse Analytics, and Azure Virtual Machines. It is intuitive for beginners and experts alike and helps companies scale experimentation efficiently and affordably.

Trial Run supports the entire experimentation lifecycle, which includes:

Trail Run illustration

Design: Build a cost-effective and efficient experiment that gives you the data you need to proceed with confidence.
Analyze: Work with variables that provide you with targeted and actionable insights.
Act: Use the generated insights to ensure your new rollout provides your stakeholders with the precise ROI.

Trial Run offers valuable support across various operational and administrative departments, including Retail, Consumer Packaged Goods (CPG), and Telecommunications.

Through its scientific and methodical testing approach, Trial Run can uncover fresh perspectives and guide decision-making through a range of tests, including:

Marketing and merchandising strategies.
Enhancing the in-store experience.
Examining store operations and processes.

These tests are carried out at the store operations and process, product, or consumer levels.

Trial Run offers a dynamic, affordable, and modern way of experimentation so you can stay relevant in a rapidly changing business environment. Trial Run also helps you to drive experiments through:

Driver Analysis: Identify key factors that are significant in driving the business outcomes
Rollout simulator: Maximize the ROI of a campaign
Synthetic Control Algorithm: Determine the right number of control stores with appropriate weights to create the replica of the test store
Experiment calendar: Avoid overlaps in experiments
Clean search: Let Trial Run parse the experiment repository and find entities that are available for a test

What you can expect from Trial Run

Graphical design elements make it easy to use the program as an expert or a beginner
Automated workflows can guide you through the process from start to finish
Highly accurate synthetic control results with automated matching processes that only require minimal human intervention
Experiments at speed and scale without the hassle of expert teams or expensive bespoke solutions
Training, troubleshooting, and best practices from the best in the business
Easy pilots to help your new idea go live in as little as 6 to 8 weeks

How do I get started?

Are you ready to implement cutting-edge technology to help you build cost-effective and efficient experiments that provide you with the data you need to make decisions?

If you want to achieve successful Trial Run implementation, get started on Azure Marketplace.

Interested in learning more about how Fractal can help you implement Trial Run, contact us to get in touch with one of our experts.

How businesses are using the metaverse and digital twins

What is the metaverse?

The metaverse has been a trending topic recently. Although there may be some variations in its definition, there are a few key elements that are the foundation of what the (or “a”) metaverse is. Whether we are referring to a metaverse for consumers or for B2B, the metaverse refers to an interconnected network of connected, persistent, and virtual worlds (or spaces) where digital avatars of human users can interact with each other and with elements of the metaverse itself. With the development of virtual reality headsets and projecting developments, the metaverse market is poised to significantly expand.

Though the recent hype about the consumer metaverse has focused on creating a social virtual space and gaming environment, the “industrial metaverse” has already taken its leap.

The Metaverse represents a global 3D network of virtual world. The transformation of the web into Web3.0 and the development of blockchain and decentralization help in creating digital assets like NFTs as digital currency using which one can purchase or sell virtual goods.

The 3 key pillars of the metaverse are:

- Virtual reality (VR)
- Augmented reality (AR)
- 3D visualization of data

Here virtual reality is computer computer-generated 3D environment that surrounds the user and responds to an individual’s action in a natural way through a head-mounted display such as Oculus Quest. On the other hand, augmented reality is the real-time use of information such as text, graphics, audio, video with real-world objects. “Pokémon Go” is a perfect example of augmented reality where it enables the user to experience the presence of a virtual world. The new Apple Vision Pro headset combines these pillars by allowing users to experience augmented reality and 3D visualization of data in an immersive environment.

Emerging metaverse technologies

Looking forward to the development of metaverse technology, many organizations are extending their capabilities. Microsoft hopes to acquire Activision Blizzard, a video game developer and publisher. Also, Unity software has acquired Weta Digital (a digital VFX company) to develop RT3D (real-time 3D). Similarly, NVIDIA is developing the Omniverse to provide the tools metaverse developers need.

To help accelerate both its first-party and third-party metaverse ecosystem, Microsoft shared what it believes are the key technology stack elements required. More than a brand-new platform, this approach builds on existing and expanding platforms and tools to provide a fairly comprehensive solution to build future B2B metaverses.

Microsoft-metaverse-technology-stack

Source: Microsoft Metaverse technology stack

Metaverse platforms

In addition to Microsoft, most technology companies, and some more traditional ones, have started to build metaverse platforms or integrations:

Trending metaverse platforms

The industrial metaverse

During the Industrial Metaverse event hosted by RealWear, Brian Vogelsang, Senior Director of Product Management at Qualcomm said: “Industrial metaverse is not a new technology. It is an evolution of technology, a convergence.”

Digital twins are representations of an intended or actual real-world physical product, system, or process. With digital twin tools such as Azure Digital Twin and 3D visualization tools such as “AltSpaceVR” and “Azure Digital Twins 3D Scenes Studio”, it is possible to visualize real-life process data in a 3D space. It helps link together 3D content, digital twin, and business-specific logic.

A great example of such use of the industrial metaverse with digital twin is the following Contoso Refinery demo from Microsoft.

Microsoft-3D-Scenes-Studio-preview-for-Azure-Digital-Twins

Source: Microsoft 3D Scenes Studio preview for Azure Digital Twins

Contoso Refinery is an end-to-end demonstration implemented with Azure Digital Twins 3D Scenes Studio. In this demo, users can monitor and diagnose real-time operational data with the visual context of 3D assets. Visualized 3D models can be easily accessed through any web browser. The demo offers a low code builder to add 3D elements to digital twins and define both user interface and business logic visualization.

Real-world use cases for the industrial metaverse

A growing number of companies have already developed industrial metaverses. Here are a few examples across different industries and use cases.

Kawasaki: Robot care and maintenance

Kawasaki unveiled the “Kawasaki DX.” It uses Microsoft’s industrial metaverse technology to implement a real-time virtual approach to robot care and maintenance. It enables remote robot operation by using a digital twin with Microsoft Azure cloud computing platform, and Microsoft’s HoloLens mixed reality.

Bosch: Integrated asset performance management

Bosch has developed a solution for asset performance management with digital twin on Azure which empowers rotating machines.

This platform enables propellers, turbines, and electric motors to indicate when they need maintenance to run with optimal costs and maximum efficiency.

Novo Nordisk: Efficiency and quality assurance

Denmark-based Novo Nordisk, a leader in the global production of medicines, is using HoloLens 2 and Dynamics 365, among others, to increase production process efficiency while maintaining the highest quality standards and acting in accordance with required requirements and regulations.

Here are a few other industrial metaverse use cases

Using AR/VR at workstation to effectively monitor and repair faults using better insights
Creating and deploying simulations suggests improvements and QA testing of workflow before physical deployments
Fast and more effective skill training. For example, in aviation, by simulating practice and using scenario-based exercises
Creating an interactive 3D space for customer engagement in retail stores

Conclusion

Use cases help in understanding the metaverse. Along with this, the available technologies help in visualizing the opportunities in the improvement of existing solutions or attempting a new venture. Industries are already testing 3D visualization of physical entities and real-life data virtual representation.

With more sophisticated hardware, the Metaverse will open new significant improvement opportunities such as testing and building products virtually before doing it in real life, it can save costs and time, test products with different conditions, and provide risk-free environment for humans.

Fractal offers end-to-end technological capabilities required for planning, building, and operating an effective industrial metaverse. Fractal has provided solutions to integrate IoT devices with the Azure cloud.

We can efficiently process and store data for effective monitoring, analysis, and visualization of the result to enhance the manufacturing value chain and effective management of resources. Feel free to contact us, if you are interested in learning more about this topic.

Databricks Spark jobs optimization techniques: Pandas UDF

Pandas UDF was introduced in Spark 2.3 and continues to be a useful technique for optimizing Spark jobs in Databricks. The idea of Pandas UDF is to narrow the gap between processing big data using Spark and developing in Python. When Spark engineers develop in Databricks, they use Spark DataFrame API to process or transform big data which are native Spark functions.

However, Python has become the default language for data scientists to build ML models, where a huge number of toolkits and libraries can be very useful. For example, while developing ML models, developers may depend on certain libraries available in Python that are not supported by Spark natively (like the basic Scikit learn library, which cannot be applied to Spark DataFrame). However, if developers develop in pure Python on Databricks, they barely take advantage of features (especially parallel processing for big data) from Spark.

In that case, Pandas UDF is there to apply Python functions directly on Spark DataFrame which allows engineers or scientists to develop in pure Python and still take advantage of Spark’s parallel processing features at the same time.

UDF vs Pandas UDF

UDF is an abbreviation of “user defined function” in Spark. Generally, all Spark-native functions applied on Spark DataFrame are vectorized, which takes advantage of Spark’s parallel processing. Although Spark already supports plenty of mainstream functions that cover most of the use cases, we might still want to build customized functions to transform data for migration existing scripts or for developers who are not familiar with Spark.

For example, let’s say we need a function to hash columns. Spark supports sha or md5 function natively, but UDF allows us to reuse the same hash and salt method on multiple columns. In addition, UDF allows the user to develop more complicated hash functions in pure Python or reuse the same function they have already developed. By converting UDF in Pandas UDF, the Pandas UDF will also process the column parallelly, which provides better performance than a UDF.

Native Spark Function

Databricks-Spark-tutorial-3-Native-Spark-Function

Spark Native Function:

11.11 seconds
Always the fastest if functions are supported

UDF

Databricks-Spark-tutorial-3-UDF

UDF:

31.84 seconds
Easy to migrate. Much slower.

Pandas UDF

Databricks-Spark-tutorial-3-PandasUDF

Pandas UDF:

24.39 seconds
Faster than UDF

Spark native functions will always have the best performance overall. However, when we have to develop some transformation function that is not supported by Spark, or it’s easier for developers to develop in pure Python, using Pandas UDF can optimize Spark jobs performance.

Grouped Map UDFs

Another useful feature of Pandas UDF is the grouped map. The grouped map feature will split a Spark DataFrame into groups based on the groupby condition, and applies user-defined function to each group, which could transform each group of data parallelly like a native Spark function.

One useful scenario for grouped map is to train multiple models based on groups when we have a training function. In pure Python, without additional parallel or groupby settings, developers will prepare a training dataset and a testing dataset for each group, then train the model one by one. By using Grouped Map UDFs, developers can apply the function on each group simultaneously, which works like parallel processing.

Sequential train

Databricks-Spark-tutorial-3-Sequential-train

Sequential train:

27.4 minutes

Apply function on each group sequentially

Spark Grouped Map Pandas UDF

Databricks-Spark-tutorial-3-Spark-Grouped-Map-PandasUDF

Spark Grouped Map Pandas UDF:

3.84 minutes
Apply Pandas UDF on each group simultaneously

There are 8 Spark executors in the cluster. After applying Pandas UDF, the performance is almost optimized 8x, which means the 8 groups are trained at the same time. The largest benefit for Grouped Map Pandas UDF is that it can be easily converted from a normal Python function. In addition, it can be applied directly to Spark DataFrame without converting into Pandas DataFrame.

Additional: Koalas

In addition to Pandas UDF, Spark org released a new package called Koalas which is also targeted to optimize Python in Spark environments. Besides using Spark DataFrame API, users can also develop functions in pure Python using Pandas API but also take advantage of Spark parallel processing.

To put it in context of Pandas UDF: Koalas can apply functions on Pandas DataFrame while Pandas UDF applies functions on Spark DataFrame.

In summary, we have three options

1. Spark DataFrame API
2. Pandas UDF on Spark DataFrame
3. Koalas API (currently Spark Pandas API) on Pandas DataFrame

All three will take advantage of parallel processing.

Looking for more Databricks Spark job optimization tutorials?

Check out some of the other techniques we’ve covered below:

Databricks Spark jobs optimization techniques: Multi-threading

Spark is known for its parallel processing, which means a data frame or a resilient distributed dataset (RDD) is being distributed across the worker nodes to gain maximum performance while processing. However, one problem we could face while running Spark jobs in Databricks is this: How do we process multiple data frames or notebooks at the same time (multi-threading)?

The benefits of parallel running are obvious: We can run the end-to-end pipeline faster, reduce the code deployed, and maximize cluster utilization to save costs. Let’s see what this looks like with an example comparing sequential loading and multi-threading.

Sequential loading vs. Multi-threading

The following scenario shows an example when we have multiple sources to read from, coalesce into one parquet file, and then write in the destination location for each part. In this scenario, coalescing into one partition can only work on one CPU core in Spark, so all the other cores will become idle. By using a multi-threading pool, each CPU will have jobs to work on, which not only saves time but also creates a better load balance.

Our test cluster has one 4 cores/8 GB master node with two 4 cores/8 GB worker nodes.

Sequential loading

Without multi-threading, under the sequential method, we read each part from the source, filter the data frame and write the result as one parquet file in the destination, which took about 20 seconds to load 8 tables.

Multi-threading

Multi-threading-pool-Databricks-tutorial

Under the same functions, after applying ThreadPool (8 threads at the same time), 8 tables can be loaded within 5 seconds which is 4x faster than the sequential loading method.

Conclusion

Multi-threading is relatively quick to set up compared with other optimization methods. The improvement could be unlimited if we have a large enough cluster and plenty of jobs to run parallelly (under suitable scenarios). We can also use the multi-threading pool to parallel run multiple notebooks which do not have dependencies on each other even if we do not have the same scenario as shown above.

The purpose of using multi-threading is not only to save time, but also to fully utilize the clusters’ compute power to save cost by finishing the same amount of jobs within less time, or within the same amount of time on a smaller cluster, which gives us more options to manage the end-to-end pipeline.

Possible scenarios for the multi-threading technique

Optimize bottlenecks in a pipeline to save end-to-end running time
Parallel run independent notebooks to optimize load balance, saving both time and cost
Read/write data from/to multiple tables

Extras

A multi-threading pool can also be developed by the “concurrent.futures.ThreadPoolExecutor” library in Python or the “scala.concurrent.ExecutionContext” library in Scala.

Want to learn more about Databricks Spark job optimization? Check out our previous blog on the topic to learn about the shuffle partition technique.

Databricks Spark jobs optimization techniques: Shuffle partition technique

Programmer writing a code on his desktop

Generally speaking, partitions are subsets of a file in memory or storage. However, Spark partitions have more usages than a subset compared to the SQL database or HIVE system. Spark will use the partitions to parallel run the jobs to gain maximum performance. While we operate Spark DataFrame, there are majorly three places Spark uses partitions which are input, output, and shuffle.

Input and output partitions could be easier to control by setting the maxPartitionBytes, coalesce to shrink, repartition to increasing partitions, or even set maxRecordsPerFile, but shuffle partition whose default number is 200 does not fit the usage scenarios most of the time. This blog will introduce general ideas about how to set up the right shuffle partition number and the impact of shuffle partitions on Spark jobs.

Key points for optimizing performance with the shuffle partition technique

Each partition size should be smaller than 200 MB to gain optimized performance.
Usually, the number of partitions should be 1x to 4x of the number of cores you have to gain optimized performance (which means creating a cluster that matches your data scale is also important).

Best practices for common scenarios

The limited size of cluster working with small DataFrame: set the number of shuffle partitions to 1x or 2x the number of cores you have. (each partition should be less than 200 MB to gain better performance)

e.g. input size: 2 GB with 20 cores, set shuffle partitions to 20 or 40

The limited size of clusters, but working with huge DataFrame: set the number of shuffle partitions to Input Data Size / Partition Size (<= 200mb per partition), even better to be the multiple of the number of cores you have

e.g. input size: 20 GB with 40 cores, set shuffle partitions to 120 or 160 (3x to 4x of the cores & make each partition less than 200 MB)

Powerful clusters which have more number of cores than the number calculated above: set the number of shuffle partitions to 1x or 2x the number of cores

e.g. input size: 80 GB with 400 cores, set shuffle partitions to 400 or 800.

Here is an example of how to improve the performance by simply changing the number of partitions on a small DataFrame working with a limited size of cluster (8 cores total).

Default 200 shuffle partitions

default-200-shuffle-partitions details-for-default-results

200 is way too much for this size of data and size of a cluster. It takes longer to allocate the jobs to finish all 200 jobs.

8 shuffle partitions to match the number of cores

8-shuffle-partitions-to-match-number-of-cores details-for-shuffle-partition-results

By simply changing the # of shuffle partitions without changing anything else, the process is running about 40% faster than the default.

Conclusion

The first and most important thing you need to check while optimizing Spark jobs is to set up the correct number of shuffle partitions. The number of shuffle partitions will not only solve most of the problems but also it is the fastest way to optimize your pipeline without changing any logic.

Note:

The example was using a small DataFrame with a limited cluster, which does not need to consider the size of each partition and has no skew keys. While optimizing a larger DataFrame, the solution will also include checking the size for each partition and making sure each partition is well distributed.

The ideal size of each partition is around 100-200 MB. The smaller size of partitions will increase the parallel running jobs, which can improve performance, but too small of a partition will cause overhead and increase the GC time. Larger partitions will decrease the number of jobs running parallel and leave some cores ideal by having no jobs to do. If you also have a skew key issue, try to add a dummy column and force Spark to partition on the well-distributed dummy column while partitioning then drop the dummy column while writing.

Check out this latest tutorial: Databricks Spark jobs optimization techniques: Multi-threading .

AI predictions for 2024: The top 7 trends to watch out for

AI is transforming the world in ways we never imagined. From creating art and entertainment to enhancing productivity and security, AI is reshaping every aspect of our lives and society.

In this blog post, we will highlight seven predictions for AI in 2024, based on the insightful LinkedIn post by our Co-founder, Group Chief Executive & Vice-Chairman, Srikanth Velamakanni.

“2023 was the year when the world woke up to the potential of AI, 2024 will be a year when we see a very tangible role of AI in making our lives better,” Srikanth said.

Now, let’s jump into some of the most exciting and impactful developments in AI and how they will affect us soon.

The foundation models race will heat up: Gemini will launch before February, followed quickly by GPT-5 by March.
Foundation models vs domain-specific models: Foundation models will continue to challenge domain-specific ones, yet hybrid approaches will remain valuable.
Productivity boost across sectors: AI assistants, apps, and agents will enhance productivity in repetitive, cognitive, and creative tasks.
Enterprise adoption of Gen AI: Over 80% of Fortune 100 companies will implement Gen AI infused use cases, as risks related to privacy and compliance get mitigated.
Enhanced enterprise and website search: Powered by Gen AI, enterprise search and website search experience will improve substantially.
AI media creation: Text-to-video, image upscaling, and image-to-video technologies will become mainstream, revolutionizing content creation.
AI assistants in coding: Near-universal adoption of these tools will boost productivity by approximately 30% in coding tasks.

If you want to know more about the other predictions, click here, where you can find the original post and join the discussion.

AI predictions for 2024: The top 7 trends to watch out for

AI is transforming the world in ways we never imagined. From creating art and entertainment to enhancing productivity and security, AI is reshaping every aspect of our lives and society.

In this blog post, we will highlight seven predictions for AI in 2024, based on the insightful LinkedIn post by our Co-founder, Group Chief Executive & Vice-Chairman, Srikanth Velamakanni.

“2023 was the year when the world woke up to the potential of AI, 2024 will be a year when we see a very tangible role of AI in making our lives better,” Srikanth said.

Now, let’s jump into some of the most exciting and impactful developments in AI and how they will affect us soon.

The foundation models race will heat up: Gemini will launch before February, followed quickly by GPT-5 by March.
Foundation models vs domain-specific models: Foundation models will continue to challenge domain-specific ones, yet hybrid approaches will remain valuable.
Productivity boost across sectors: AI assistants, apps, and agents will enhance productivity in repetitive, cognitive, and creative tasks.
Enterprise adoption of Gen AI: Over 80% of Fortune 100 companies will implement Gen AI infused use cases, as risks related to privacy and compliance get mitigated.
Enhanced enterprise and website search: Powered by Gen AI, enterprise search and website search experience will improve substantially.
AI media creation: Text-to-video, image upscaling, and image-to-video technologies will become mainstream, revolutionizing content creation.
AI assistants in coding: Near-universal adoption of these tools will boost productivity by approximately 30% in coding tasks.

If you want to know more about the other predictions, click here, where you can find the original post and join the discussion.

AI predictions for 2024: The top 7 trends to watch out for

AI is transforming the world in ways we never imagined. From creating art and entertainment to enhancing productivity and security, AI is reshaping every aspect of our lives and society.

In this blog post, we will highlight seven predictions for AI in 2024, based on the insightful LinkedIn post by our Co-founder, Group Chief Executive & Vice-Chairman, Srikanth Velamakanni.

“2023 was the year when the world woke up to the potential of AI, 2024 will be a year when we see a very tangible role of AI in making our lives better,” Srikanth said.

Now, let’s jump into some of the most exciting and impactful developments in AI and how they will affect us soon.

The foundation models race will heat up: Gemini will launch before February, followed quickly by GPT-5 by March.
Foundation models vs domain-specific models: Foundation models will continue to challenge domain-specific ones, yet hybrid approaches will remain valuable.
Productivity boost across sectors: AI assistants, apps, and agents will enhance productivity in repetitive, cognitive, and creative tasks.
Enterprise adoption of Gen AI: Over 80% of Fortune 100 companies will implement Gen AI infused use cases, as risks related to privacy and compliance get mitigated.
Enhanced enterprise and website search: Powered by Gen AI, enterprise search and website search experience will improve substantially.
AI media creation: Text-to-video, image upscaling, and image-to-video technologies will become mainstream, revolutionizing content creation.
AI assistants in coding: Near-universal adoption of these tools will boost productivity by approximately 30% in coding tasks.

If you want to know more about the other predictions, click here, where you can find the original post and join the discussion.

4 steps to improve the data quality of your organization

The world today is swimming with data, and organizations need to handle a large amount of it to derive valuable insights for decision-making. But data is only helpful if it is of good quality. According to SAP, bad data costs the US $3 trillion annually. These costs include employee’s time and effort to correct bad data and errors.

Why is data quality important?

Data quality is a base for data analytics and data science. It measures how well-suited your data is to accomplish a particular task accurately and consistently. Good data helps an organization make key spending decisions, improve operations, and develop growth tactics. Even though technologies like AI and machine learning have enormous potential to handle large volumes of data, they need good quality data to produce reliable results quickly.

As data is an integral part of an organization, data quality impacts many aspects, from marketing to sales to content creation.

Good quality data helps you make informed decisions, target the right audience, drive effective marketing campaigns, strengthen customer relationships, gain competitive advantage, and so on.

Benefits of improving data quality

In this competitive era, organizations try to understand their customers better and make better financial, marketing, and development decisions based on accurate data for a better ROI. Bad data is unstructured data that may show quality issues like inconsistency, inaccuracy, insufficiency, or even duplicate information. It could be misleading and even more harmful for a business than a lack of data.

Improving the data quality of your company can result in the following advantages:

Benefits of improving data quality

Data-driven decision-making: Decision-making is based on solid reasoning, and the correct data can only help make wise business decisions and provide the best outcomes.
Customer intimacy: Drive marketing and customer experience by analyzing entire consumer views of transactions, sentiments, and interactions by using data from the system of record.
Innovation leadership: Learn more about your products, services, usage trends, industry trends, and competition outcomes to help you make better decisions about new products, services, and pricing.
Operational excellence: Make sure the correct solution is provided quickly and dependably to the right people at a fair price.

Challenges faced while maintaining data quality

Poor data quality can lead to financial losses, increased complexity, and limited usefulness of real-world evidence. Understanding data quality challenges is crucial for informed decision-making, reducing business risks, and achieving data-driven goals. Explaining a few challenges below.

Data debt reduces ROI: Data debt refers to the negative consequences of accumulating poorly managed data. This includes inaccurate, inconsistent, incomplete, or inaccessible data. If your organization has a lot of unresolved data-related problems, such as issues with data quality, categorization, and security, it can hinder you from achieving the desired Return on Investment (ROI).
Lack of trust leads to lack of usage: A lack of data confidence in your organization leads to a lack of data consumption, which has a detrimental impact on strategic planning, KPIs, and business outcomes.
Strategic assets become liabilities: Poor data puts your company in danger of failing to meet compliance standards, resulting in millions of dollars in fines.
Increased expenses and inefficiency: Time spent correcting inaccurate data equals less workload capacity for essential efforts and an inability to make data-driven decisions.
Adoption of data-driven technologies: Predictive analytics and artificial intelligence, for example, rely on high-quality data. Delays or a lack of ROI will come from inaccurate, incomplete, or irrelevant data.
Customer experience: Using bad data to run your business can hinder your ability to deliver to your customers, increasing their frustration and reducing your capacity to retain them.

Improving data quality with Fractal’s methodology

Maintaining high levels of data quality allows organizations to lower the expense of finding and resolving incorrect data in their systems. Companies can also avoid operational errors and business process failures, raising operating costs and diminishing revenues.

Good data quality enhances the accuracy of analytics applications, leading to improved business decisions that increase sales, improve internal processes, and provide firms with a competitive advantage over competitors. High-quality data can also help increase the use of BI dashboards and analytics tools. If analytics data is perceived as trustworthy, business users are more inclined to depend on it instead of making judgments based on gut feelings or their spreadsheets.

Fractal has developed a process that significantly improves data quality across enterprises. Here’s how we do it.

Data Preparation and Rule Calculations

Fractal helps handle large volumes of data, preparing it for further processing. It also performs calculations based on data rules, identifying defective records in the data.

Data extraction and preparation

Fractal leverages a back-end engine for data extraction, data preparation, and data rules configuration on various data sets.

Optimized process

The focus is an optimized process with minimal processing time, parallel processing, and data aggregations to reduce the storage space and provide the user’s best dashboard performance.

Data quality improvement

Fractal helps transform the data cleansing operation with faster reduction of data defects, improved data quality, and tracking key KPIs like asset score, coverage, and conformance.

How can Fractal help maintain data quality with Google Cloud?

Fractal leverages all the services provided by Google Cloud and supports integrations with the Google Cloud Platform. It also determines which Google Cloud services best meet Fractal’s data quality needs for each project.

Here are some ways Google Cloud can help maintain data quality.

Data Governance: It helps automatically detect and protect sensitive information in data, ensuring data privacy and compliance. It also helps enable granular control over data access, preventing unauthorized modifications or deletions.
Data Profiling & Cleansing: It offers a user-friendly interface for data cleaning and transformation, including outlier detection, data standardization, and data validation. It also provides AI-powered tools for data profiling and anomaly detection, helping identify data quality issues proactively.
Data Monitoring & Improvement: It offers comprehensive dashboards and alerts for monitoring data quality metrics, allowing for proactive identification and resolution of issues. It also helps run large-scale data quality checks and analysis, providing deeper insights into data quality trends.
Machine Learning & AI Integration: It provides tools for developing and deploying custom AI models for data quality tasks, such as entity extraction, classification, and matching. It also helps build serverless data quality functions that can be triggered on specific data events, ensuring real-time data quality checks.

Conclusion

In today’s data-driven world, maintaining high-quality data is no longer an option but a necessity. Poor data quality can lead to financial losses, operational inefficiencies, and inaccurate decision-making. By leveraging Fractal’s data quality methodology and Google Cloud’s powerful tools and services, organizations can effectively address data quality challenges and unlock their full potential.

Fractal empowers organizations to achieve data quality excellence. When combined with Google Cloud’s data quality capabilities, Fractal delivers a comprehensive solution for managing and improving data quality throughout the entire data lifecycle.

Are you seeking help to improve the data quality of your organization? Contact us to get started!

Top 7 Announcements from AWS re:Invent 2023

Amazon Web Services recently concluded its highly anticipated re:Invent 2023 event, showcasing a resurgence of big community events in the tech industry after the pandemic-related hiatus. With a record-breaking attendance of around 50,000 participants, re:Invent 2023 was a milestone event with significant announcements and insights for the tech world.

Here’s a summary of top announcements, followed by the keynote from Adam Selipsky, AWS CEO, and in-depth announcements.

Generative AI was, unsurprisingly, the buzzword. AWS rolled out an exciting new AI chip, support for new foundation models, updates to its generative AI platform Amazon Bedrock, and support for vector databases and zero-ETL integrations. They also announced a new generative AI assistant dubbed Amazon Q (in reference to Star Trek’s “Q” character, seemingly). Q will probably have a broad set of use cases in enterprises.

1. Amazon Q: A new generative AI assistant

A Gen AI-powered assistant designed to help get relevant answers, solve problems, and generate content using Retrieval Augmented Generation (RAG). It also integrates with other AWS services:

AWS Console: Amazon Q simplifies exploring AWS’s framework, best practices, and documentation. It is accessible in the AWS management console. For example, “How to build a web application on AWS?” yields a list of services like AWS Amplify, AWS Lambda, and Amazon EC2, along with their advantages and resources.
Connect: Cloud-based contact center service ensures scalable customer service operations. Amazon Q enhances customer interactions by understanding needs, minimizing wait times, and delivering real-time responses via API integration.
Amazon QuickSight: Business intelligence service offering interactive dashboards, paginated reports, and embedded analytics. Amazon Q within QuickSight enhances productivity for business analysts and users by swiftly creating visuals, summarizing insights, answering data queries, and constructing data stories using natural language.

2. Expanded choice of models in Amazon Bedrock

Bedrock is the foundation model library from AWS. It enables the integration of various models that underpin generative AI. AWS have introduced some new models to Amazon Bedrock:

Meta’s popular LLaMA-2
Anthropic Claude 2.1: An update to Claude 2, it now offers a 200k token context window (vs. 128k for GPT 4 Turbo), reduced rates of hallucination, and improved accuracy over long documents.
Amazon Titan Image Generator: Similarly to tools such as Mid-journey, Stable Diffusion, and Dall-E, Titan Image Generator lets you not only create but also improve images with natural language commands. Titan also supports enterprise needs for invisible watermarks on images.
Amazon Titan Multimodal Embeddings: Improve searches by understanding images and text. For instance, a stock photo company could use it to find specific images based on descriptions or other images, enhancing accuracy and speed.

3. Four New Capabilities for AWS Supply Chain

An application that unifies data and provides ML-powered actionable insights. It incorporates embedded contextual collaboration and demand planning features while seamlessly integrating with your client’s current enterprise resource planning (ERP) and supply chain management systems. Announced three new capabilities:

Supply planning (Preview): Plans purchases of raw materials, components, and finished goods. This capability considers economic factors, such as holding and liquidation costs.
Visibility and sustainability (Preview): Extends visibility and insights beyond your client’s organization to your external trading partners. This visibility lets you align and confirm orders with suppliers, improving the accuracy of planning and execution processes.
Amazon Q (Preview): As mentioned above, Amazon Q is now integrated with Supply Chain services to empower inventory managers and planners with intelligent insights and explore what-if scenarios.

4. SageMaker capabilities (Preview)

HyperPod: accelerates model training by up to 40%, enabling parallel processing for improved performance.
Inference: reduces deployment costs and latency by deploying multiple models to the same AWS instance.
Clarify: supports responsible AI use by evaluating and comparing models based on chosen parameters.
Canvas: enhancements facilitate seamless integration of generative AI into workflows.

5. Next-generation AWS-designed chips (Preview)

Amazon jumped into the custom cloud-optimized chip fray and introduced two new chips:

Graviton4: A powerful and energy-efficient chip suitable for various cloud tasks.
Trainium2: A high-performance computing chip, it helps accelerate model training while making it more cost-effective.

Notable AWS customers like Anthropic, Databricks, Datadog, Epic, Honeycomb, and SAP already use these chips.

6. Amazon S3 Express One Zone

A new purpose-built storage class for running applications that require extremely fast data access.

7. New integrations for a zero-ETL future

Aim to make data access and analysis faster and easier across data stores. Four new integrations are:

Amazon Aurora PostgreSQL, Amazon DynamoDB, and Amazon RDS for MySQL integrations with Amazon Redshift: Query and analyze data from multiple relational and non-relational databases in Amazon Redshift without building and maintaining custom data pipelines.
Amazon DynamoDB integration with Amazon OpenSearch Service: Perform full-text and vector search on DynamoDB data in near real-time.

To learn more about all those exciting news, please check the AWS Re:Invent keynote blog. 

Fractal enables Machine Vision over Azure 5GPMEC with new Vision Analytics as a Service offering

It is an exciting time in AI (Artificial Intelligence). ChatGPT is revolutionizing the world, and one of the best-known secrets of Microsoft’s platform is set to revolutionize the world of vision AI processing, through a deep technical collaboration with NVIDIA & Microsoft.

Machine vision cameras

Today, machine vision cameras are commonly deployed worldwide for visual inspections. These cameras enable people to see what they cannot see with the naked eye and have many use cases. They are frequently used in manufacturing settings for quality control on production lines and in detail-required industrial scenarios. Basler & Allied vision are leading producers of these cameras, which operate against the Gig-E vision protocol. NVIDIA announced compatibility and substantial investment in this protocol at GTC in March. Today, these cameras can be used to identify quality control issues and stitched together into any solution supporting and running with DeepStream.

Machine vision cameras

The importance of hardware acceleration

However, these cameras come with extremely specific requirements. Hardware that can process multiple streams of high-fidelity video coming from high frame rate cameras must be able to provide the right form factor, operating environment constraints, and processing requirements. Fortunately, NVIDIA’s GPUs (graphics processing units) can be leveraged for parallel processing at scale. This idea of massively parallelized hardware acceleration is in testing today for deployment on Azure Stack Edge and other Azure Certified Devices and will enable at-scale GPU deployment for hardware-accelerated machine vision running at the edge and connected to the cloud.

Edge and cloud

Finally, Azure’s Cognitive Services team’s unique and innovative Florence models (in preview today) enable customers to rapidly create and innovate upon cloud models – to accelerate edge-to-cloud hybrid AI processing workloads development. Azure’s Spatial Analysis cognitive service already supports edge deployments of certain industrial use cases on Azure Stack Edge. Leveraging these tools also allows for both edge deployments very rapidly and for hybrid deployments – leveraging custom models at the edge to identify and pass relevant frames to the cloud for deeper analysis.

Introducing Vision as a Service

Fractal has created Vison Analytics as a Service, an offering based on the repeated implementations of cloud-connected edge intelligence deployed by their Edge & IOT team. This stitches together the innovations offered by Microsoft and NVIDIA through a Managed Application, available to customers for a simple price. This simple managed offering allows for unified and integrated pricing as per the scale factors & standard requirements of a complex vision implementation, with an initial implementation of vision AI deployed in a 3-month timeline. It leverages standard templates and accelerators from Microsoft and NVIDIA, like the Azure DeepStream Accelerator, the NVIDIA GPU sharing work, MEC app solution accelerator, and the innovations mentioned above to enable a rapid deployment package for customers wanting to implement video AI solutions at scale.

This offering also provides several optional industrial extensions, including camera management, coordination with sophisticated AI engines, and integration with critical industrial equipment assets. One common use case is the deployment of deep reinforcement learning to fully automate command and control by leveraging a physics-based simulation on a manufacturing line, to drive activity after visual inspection. It can also be integrated with PLC control systems and other required industrial assets, as a direct extension, making it an incredibly useful solution for organizations wishing to upgrade their quality control processes with vision AI.

The goal with this offering is to enable as many customers as possible to unlock the value of vision AI by leveraging Fractals capabilities with the greatest ease possible.

We also offer a free consultation to scope a vision AI engagement and identify a hardware recommendation, edge & cloud strategy, audit camera requirements, scope the required skills, and any required DRL (Deep Reinforcement Learning) deployments. This standardized pricing is designed to offer greater scale, while Fractal leverages vision intellectual property development alongside Microsoft and NVIDIA’s offerings, enabled by this offer.

Transform Customer Digital Experience with AIDE

High digital abandonment rates are typical for brands across domains, driven mainly by the experiential issues site users face during their journey. Identifying these friction points can be burdensome for businesses as they need help digesting the new wealth of granular data generated along their customer’s digital journey.

Fractal’s Automated Insights for Digital Innovation (AIDE) is a smart digital solution for a broad range of industries, including retail, finance, insurance, and more, that uses a customizable open-source AI that works well with complex journeys, data security, and time-to-market needs for multiple industries. It helps make smarter decisions at every step and enables businesses to resolve issues quickly and increase potential revenue and leads.

AIDE helps:

Analyze millions of digital consumer touchpoints to provide insights to increase revenue, engagement, and growth for the business.
Identify the root cause of friction from call, chat, and website errors and use AI to parse out critical signals from all the unstructured data in the customer journey.
Get the most comprehensive insights into the digital journey to drive data-driven hypotheses for supporting A/B tests to drive website design changes to improve the consumer experience.

What is AIDE?

AIDE is a digital optimization platform that helps detect and contextualize the issues faced by visitors on digital channels. It acts as an intelligent digital solution for various industries, including retail, finance and insurance, telecommunications, tech, media, and more. AIDE uses customizable, open-source AI that works well with complex journeys, data security, and time-to-market needs for multiple industries. It’s an insight-orchestrated platform supported by natural language-generated insights. AIDE:

Selects the sales or service flow to influence particular focus points.
Identifies the selected data domains to create a journey sensor.
Helps detect the most important anomalies across key performance indicators.
Finds the friction point on the website using various journey sensors.
Helps analyze the customer voice to add context to insights.

Leveraging the power of the AWS cloud, AIDE is built on Amazon RDS, Redshift, EMR, LaMDA, E2, S3 and can be deployed in your AWS environment.

How can AIDE help my business?

AIDE product architecture

AIDE brings together data engineering, Natural Language Processing (NLP), machine learning, and UI capabilities to help clients:

Easily develop new data features from raw data to power downstream data analytics use cases.
Identify and locate precise points of friction on your company’s website, online events, or funnels.
Deep dive into the context of customer dissonance using the voice of customer analytics.
Prioritize the most critical areas based on value loss estimation.
Analyze Omni-channel customer journey analysis.
Provide user-friendly and intuitive UI for beginners and experts.
Provide root cause analysis of customer pain points/dissonance during the digital journey.

What can I expect from AIDE?

With AIDE, you can capture every in-page interaction and micro-gesture to understand the site user’s journey and identify frustrations and errors impacting conversion and self-serve rate.

AIDE helps companies remove friction and errors that spoil the visitor’s experiences. It also helps leverage best-in-class AI/ML modules to identify struggle points and recommend changes to drive design improvements using multiple features such as:

Sensorize: An automated AI/ML pipeline derives meaningful business indicators using the click activity across customer journeys.
Detect: Deviations from expected behavior across digital journeys get captured by applying pattern recognition algorithms to the key digital indicators.
Locate: A suite of supervised machine learning algorithms identify drivers of key customer journey outcomes (drop-off, clear cart, etc.) and measure relative impact at a page and click level of a customer’s experience.
Reveal: NLP module performs sentiment analysis and entity extraction on the voice of customer data such as chat; feedback etc. to identify the root cause of the friction and generate actionable insights.
Prioritize: Quantify the insights with respect to loss in revenue or incremental overhead costs to prioritize hypotheses for improving website design.

Overall, AIDE is adaptable and open source, making it a flexible solution for various needs and is effective at addressing the complex customer journeys of today’s digital world. It is secure and reliable, with a focus on data protection, and easy to use and deploy, with a quick time to market.

How do I get started?

AIDE is also available on AWS Marketplace. Contact us to learn how an AIDE solution can identify and reduce friction points, helping the business grow at scale.

Effective Enterprise Data Management using Federated DataOps

In the past, organizations have built monolithic analytics solutions with a high degree of centralization around a single data engineering team responsible for maintaining the platform and building data pipelines. This approach doesn’t work in large organizations as they operate in a decentralized way. That’s why new data management approaches such as data mesh have emerged to tackle conventional data management challenges such as data redundancy and dependency, privacy and security, limited data sharing and availability, and data inconsistency.

The underlying reasons for these challenges are that traditional strategies fail to address the following adequately:

Complexity: Data projects can be complex, involving multiple stakeholders, systems, and technologies. This complexity can make coordinating efforts difficult and ensure the project is delivered on time and within budget.

Lack of resources: Data projects often require specialized skills and expertise, which can be challenging to find and retain. The results can be delays and cost overruns as organizations need help to secure the necessary resources.

Data quality: Data experts are spread across the organization, departments, and business units. Centralized responsibility for managing data solutions leads to low quality. Good data quality can lead to correct insights and sound decision-making.

Lack of governance: With proper governance and controls, data projects can become more cohesive and organized, leading to delays and suboptimal outcomes.

Organizations must adopt modern data science solutions that can efficiently manage data’s ever-increasing analytics need, volume, and complexity.

Fractal recommends leveraging Federated DataOps with your organization’s data management strategies to provide accuracy, consistency, and faster time-to-value.

Federated DataOps

Data operations (DataOps) is evolving daily to meet automation needs for organizations ever-increasing data and analytics needs. DataOps aims to improve the quality, speed, and reliability of data processing and data-driven decision-making within an organization.

DataOps automates data flows between data managers and data consumers within an organization. By improving and automating these processes, DataOps can help organizations leverage their data assets faster and better and improve the quality of the insights. DataOps often uses a combination of tools and technologies, such as data integration platforms, data lakes, and data governance frameworks, to support the automation and integration of data flows across the organization.

Enterprise Federated DataOps architecture

Modern data management strategies like Data Mesh and Data Fabric architectures can use Federated DataOps Governance to help manage and automate the data flow within an organization. There are several reasons why organizations might choose to use Federated DataOps in conjunction with data fabric and data mesh architectures:

Federated governance: Organizations are adopting self-serve distributed data platforms, which could lead to silos and duplicate DataOps practices by different teams. Federated DataOps ensures high standards, reusability, and governance across teams.
Improved data quality: Continuous integration and delivery can enhance data quality by catching and fixing errors early in the data pipeline. It ensures data accuracy and reliability.
Increased speed: Automating and streamlining data integration and management processes can help organizations get insights from their data faster and make quicker decisions based on those insights.
Enhanced collaboration: It promotes collaboration and continuous improvement, fostering better communication and cooperation between data managers and data consumers within an organization. It leads to progress in the overall effectiveness of data analytics processes.
Reduced complexity: It can help reduce the complexity of data management and analytics processes by automating and streamlining the data flow within an organization, enabling quick and informed decisions.

Get started with your data management journey with Fractal

Companies are looking for a faster and more efficient way of managing data and preserving and using their data as technology advances. A sound data strategy is key to data’s successful and practical usage to achieve business goals within the stipulated time. The process demands quickness, efficiency, and consistency; hence, DataOps is used with Data Mesh or Data Fabric to accelerate the implementation of data management strategies.

Fractal Analytics is a data & AI Microsoft Solutions Partner that catalyzes data-driven transformation initiatives for enterprises leveraging the Microsoft cloud. We assist enterprises in creating end-to-end data management solutions that encourage innovation and value generation in line with current and anticipated market trends and long-term corporate objectives.

Role of data & analytics in alleviating health inequities

The COVID-19 pandemic and the resulting global recession caused a significant increase in poverty around the world as many families lost their sources of income. The poorest families, especially those without access to social protection, were hit the hardest. As a result, the World Health Organization urged governments and health organizations to pay attention to the social determinants of health to take steps to optimize community well-being.

Social determinants of health (SDOH) are any of a variety of non-medical factors that influence health outcomes. They encompass the conditions in which people are born, grow, live, learn, work, play, worship, and age — shaping their overall health. These factors include attributes associated with one’s environment, patterns of social engagement, and one’s sense of security and well-being. Collectively, SDOH influences health, working life conditions, and quality of life.

In this article, we’ll discuss how social determinants of health impact overall wellness across populations, including disparities in healthcare, and the role data can play in alleviating these inequities and shaping healthcare outcomes.

Social determinant factors

Patients’ health depends not only on the care they receive in a provider’s office but also on social determinants of health such as individual lifestyle choices, living situation, and access to healthy food.

According to a study by the CDC, social determinants of health fall into five broad groups:

HIMSS Illustration

Key benefits of studying social determinants

Addressing social determinants is important for improving health and reducing health disparities. Understanding social factors gives valuable insight into possible negative health outcomes for patients in many domains. Below are the key benefits of studying social determinants of health.

Holistic healthcare: Incorporating social determinants into healthcare practices fosters a more holistic and comprehensive approach to patient care. For instance, factors such as a patient’s education, income level, and environment should be considered when providing treatment and care.

Addressing health inequities: Social determinants have an important influence on health inequities – the unfair and avoidable differences in health status seen within and between countries.

Resource development: Acknowledging social determinants can initiate the development of resources to solve inequality and strengthen the overall health of the community.

Influencing health outcomes: Research shows that social determinants can be more important than healthcare or lifestyle choices in influencing health outcomes.

The impact of social determinants of health

Social determinants of health have a significant impact on people’s health, well-being, and quality of life. A growing body of research indicates:

Children born to parents who haven’t completed high school are more likely to live in environments that contain barriers to health.
Poor individuals who are white are less likely to live in areas of concentrated poverty than poor racial and ethnic minorities.
As income decreases, the likelihood of premature death increases.
There is a direct link between the likelihood of smoking, shorter life expectancy, and lower income.
The environment in which an individual lives may impact future generations.
Stress related to disparities has a direct link to health and often results from overlapping factors.

Negative social determinants of health can lead to disparities in healthcare, which can be costly and inhibit the overall quality of care and population health. This can result in added healthcare expenses, loss of productivity, and premature death. According to the Kaiser Family Foundation, 30% of direct medical costs for black, Hispanic, and Asian Americans are unnecessary costs incurred due to inefficiencies, disparities, or inequities in the healthcare system. In addition, the US economy loses an estimated $309 billion annually due to the direct and indirect costs of disparities.

Role of data & analytics in alleviating healthcare inequities

Barriers like a lack of data standards and costly datasets can hinder an organization’s access to social determinants information. However, developing an approach for more holistic patient care will be necessary for organizations looking to improve patient and population health, whether the data is complete or not.

Data and analytics are vital in helping to end these disparities and ensuring that all populations have the same access to services and care, not only for COVID-19 but also for all diseases and disorders that threaten public health.

Through data analytics and population health management, providers can improve patient outcomes, enhance care management, and address social determinants of health. Nowadays, data analytics are helping providers replace the “one size fits all” care mentality to deliver value-based care. Providers can assess which processes are the most effective methods for wellness and prevention within value-based care models. With population health management, organizations can consider physical and social determinants of health that may impact individuals and focus on “well care” rather than waiting for a patient to become ill.

Building a better healthcare system with Fractal

Health disparities and inequities are shaped by a multitude of factors in an individual’s socio-economic and healthcare journey. The health outcomes of an individual are significantly influenced by social determinants of health of the community in which they reside.

Fractal’s RAISE, powered by AWS, is an AI-powered solution that helps organizations speed up population health and health equity journeys. RAISE combines a member’s community data with clinical and social needs, assisting organizations in crafting the right interventions. It also creates a multifaceted approach involving data, analytics, and AI to advance health equity. It helps identify the drivers of inequities and disparities that are specific to the members of the organization.

AWS equips Fractal’s AI/ML solutions with the scalability, reliability, and security needed to deliver its solutions to healthcare providers and patients worldwide. By leveraging AWS, Fractal is helping to make a real difference in the lives of people affected by health inequities by supporting community-based healthcare organizations, advocating for policies that address the social determinants of health, and promoting health education and literacy.

Fractal recently led an on-demand webinar in collaboration with the Healthcare Information and Management Systems Society (HIMSS) to discuss health inequities. It explained how Fractal has been solving various healthcare-related issues using advanced AI/ML solutions.

The webinar featured Fractal’s Chief Practice Officer, Matt Gennone, and Dr. David Nash, an advisor at Fractal. They discussed the economic impact of health inequities and delayed care and how healthcare can return to its “true north” of providing high-quality care.

To know more, watch this session.

7 ways implementing MLOps can transform your business

Machine learning operations (MLOps)

As per Gartner, only 50% of machine learning models reach production, leading to an efficiency challenge for many organizations. To improve the rate of machine learning model deployment, many organizations have begun to adopt MLOps.

MLOps is the concept of applying DevOps principles to ML (Machine Learning) workflows and helping increase the deployment rate of ML models to production by leveraging a set of practices, processes, and tools per the used case scenario.

The MLOps process includes several stages: data preparation, model training, testing and validation, deployment, monitoring, and feedback loops. The main goal of MLOps is to increase the machine learning process’s efficiency, accuracy, and reliability while ensuring the models perform as expected in real-world environments.

MLOps typically involves using various processes, tools, and technologies, such as version control systems, containerization platforms, and continuous integration and deployment pipelines, to automate and streamline the machine learning workflow. The tools and practices enabled by MLOps can help reduce the risk of errors, improve collaboration, and ensure that machine learning models are updated and optimized for better performance over time.

Let us look into the 7 key benefits of implementing MLOps

7 key benefits of implementing MLOps-illustration

1. Increases productivity

MLOps practitioners leverage various tools and practices designed to help streamline and automate machine learning development and deployment processes. It can include automating data preprocessing and feature engineering, managing model training and evaluation, and deploying and monitoring models in production.

By implementing tools designed to automate and standardize development and deployment processes, organizations can reduce the time and effort required to develop and deploy machine learning models, allowing data scientists and engineers to focus on higher-level tasks.

This results in the faster and more efficient delivery of high-quality machine learning models, ultimately driving business value and improving productivity.

2. Faster deployment and easy monitoring

MLOps methodologies can help organizations accelerate modeling processes. They can also help facilitate machine learning models’ seamless construction and deployment by helping leverage automated systems.

Commonly used MLOps tools can help with automatic system monitoring systems, which can be used in the continuous monitoring of models in production, allowing for quick identification and resolution of any issues. These tools help organizations improve the speed and quality of their machine learning deployment, leading to increased productivity and better outcomes.

3. Budget and cost management

Implementing MLOps can help ensure efficient resource usage and cost control by using tooling designed to monitor usage patterns, identify bottlenecks, and scale resources based on demand. These tools can help estimate and track costs before, during, and after experimentation.

4. Reproducibility and versioning

Organizations can leverage MLOps policies to enable a structured approach for practitioners and data scientists to track and manage changes, enable versioning, and provide a history of edits and versions created.

Versioning aids in deploying models in production and enhances the reliability and scalability of machine-learning models.

5. Reliability

Leveraging MLOps methods and tools can result in more reliable ML pipelines by minimizing the scope of human error and providing real-time data insights. MLOps can improve the dependability of machine learning models, ensuring their consistency and accuracy in production.

By continuously monitoring model performance and dependencies, teams can promptly identify and address issues, increasing the models’ reliability.

6. Collaboration

MLOps best practices and policies are meant to break down silos between teams, allowing them to collaborate, share data more efficiently, and seamlessly integrate their workflows. This collaboration can lead to faster and more efficient model deployment and a more streamlined machine-learning process.

With MLOps, organizations can ensure that their machine learning projects are connected and working together efficiently, leading to better outcomes and improved productivity.

7. Monitorability

Through MLOps, organizations can get insights into model performance and retrain the model continuously to ensure it gives the most accurate output. MLOps enables practitioners to do this by providing guidance on best practices for successfully implementing automated monitoring systems. These monitoring systems facilitate constant model monitoring and allow stakeholders to identify any issues or anomalies that may arise quickly.

Identifying problems and irregularities can help to improve model performance and reduce downtime, leading to better outcomes and a more efficient deployment process.

With MLOps, organizations can ensure that their machine learning models always perform optimally, improving productivity and better business results.

MLOps with Fractal

Fractal has a successful track record of delivering ML projects for clients leveraging our in-house MLOps methodology. Our secret lies in using a reliable methodology designed to remove uncertainty, foster consistency, and enable the quick realization of value, as well as continuous packaging, validation, and deployment of models to production—partnering with us for MLOps implementation grants you access to this proven methodology.

Getting started with MLOps

Organizations looking to implement MLOps can leverage the same proven methodology we use in-house to deliver projects for clients successfully.

Scale AI model governance with the AI Factory framework

AI and machine learning projects are on the rise. According to Gartner, 48% of CIOs and tech executives have deployed, or plan to deploy, an AI/ML project in 2022. It is also estimated that 50% of IT leaders will struggle to drive their AI initiatives from Proof of Concept (PoC) to production through 2023.

Challenges moving AI/ML initiatives from PoC to production

What causes the gap between PoC and AI/ML model implementation?

IT and business leaders often cite challenges relating to security, privacy, integration, and data complexity as the key barriers to deploying AI/ML models in production. It is often due to governance frameworks not being shared across an organization to ensure compliance and maintainability – if a framework exists at all.

“At some point, your proof-of-concept is likely to turn into an actual product, and then your governance efforts will be playing catch-up,” writes Mike Loukides in an O’Reilly report. “It is even more dangerous when you’re relying on AI applications in production. Without formalizing some kind of AI governance, you’re less likely to know when models are becoming stale, when results are biased, or when data has been collected improperly.”

AI models require constant attention in production to achieve scalability, maintainability, and governance. To do that, organizations need a strong MLOps foundation.

Leveraging MLOps at scale

In one survey, Deloitte found that organizations that strongly followed an MLOps methodology were…

3x more likely to achieve their goals
4x more likely to feel prepared for AI-related risks
3x more confident in their ability to ethically deploy AI initiatives

AI Factory framework benefits

Organizations following an MLOps methodology also gain a clear advantage in time to deployment. McKinsey found that companies without a formalized MLOps process often took 9 months to implement a model. In comparison, companies applying MLOps could deploy models in 2 to 12 weeks!

The secret? By applying MLOps practices, these companies were able to create a “factory” approach for repeatable and scalable AI/ML model implementation. Their engineers weren’t building everything from scratch–they could pull from a library of reusable components, automate processes, and ensure compliance and governance throughout the organization.

Luckily, you can also take this approach with our AI Factory Framework.

Our AI Factory Framework

The AI Factory Framework is a cloud-based MLOps framework that provides organizations with the foundation to deliver Data Science, Machine Learning, and AI projects at scale. It offers enterprise-level reusability, security, integration, and governance.

Simply put, AI Factory helps customers scale MLOps, centralize governance, and accelerate time to deployment.

Key benefits of the AI Factory

By leveraging reusable and standardized artifacts, automated pipelines, and governance solutions, our AI Factory framework reduces duplicate effort and upskilling needs between teams and projects.

AI Factory Framework benefits

Customers leveraging the AI Factory Framework can take advantage of our AI engineering best practices to accelerate deployment and ensure model governance at scale.

AI Factory also helps businesses:

Make the entire end-to-end lifecycle more repeatable, governable, safer, and faster
Shorten planning and development with accelerated time to deployment
Streamline operational, security, and governance processes
Reduce development risks & improve model quality
Reduce team’s upskilling needs
Achieve higher success rates & ROI

Learn more

Over the last decade, we have helped many customers build and execute their AI governance strategy. We distilled this experience and the derived best practices in this framework, to help deliver customers’ AI/ML initiatives at scale.

Want to set up your own AI Factory Framework? Contact us to get in touch with one of our experts!

Resources:

Deloitte Insights – State of AI in the enterprise
Fractal’s MLOps capabilities

ML Operationalization: Building a path to real-world business success

Operationalizing and scaling machine learning to drive business value can be challenging. While many businesses have started diving into it, only 13% of data science projects actually make it to production. Moving from the academics of ML to real-world deployment is difficult, as the journey requires finetuning ML models to fit the practical needs of a business and ensuring the solution can be implemented at scale.

Many organizations struggle with ML operationalization due to a lack of data science and machine learning capabilities, difficulty harnessing best practices, and insufficient collaboration between data science and IT operations.

Common challenges with ML operationalization

Many organizations get attracted to buzzwords like “machine learning” and “AI,” and spend their development budgets pursuing technologies rather than addressing a real problem. ML projects are an investment, and obstacles in operationalizing the solution make it even harder for the business to realize value from these solutions.

Here are some common ML operationalization bottlenecks and the solutions to tackle them.

Lack of communication, collaboration, and coordination: Proper collaboration between the data scientist team and other teams, like business, engineering, and operations, is crucial. The ML project may not add real-world business value without proper alignment and feedback.
Lack of a framework or architecture: When ML models lack the proper framework and architecture to support model building, deployment, and monitoring, they fail.
Insufficient infrastructure: ML models use vast data to train the model. Most of the time is spent preparing data and dealing with quality issues without the proper infrastructure. Data security and governance are crucial factors that must be considered in the initial phase.
The trade-off between prediction accuracy and model interpretability: Complex models are generally harder to interpret but provide more accurate predictions. The business must decide what’s an acceptable tradeoff to get a “right-sized” solution.
Compliance, governance, and security: The data science team may not always consider other issues like legal, compliance, IT operations, and others that occur after the deployment of ML models. In production, setting up performance indicators and monitoring how the model can run smoothly is important. So, understanding how ML models run on production data is a crucial part of risk mitigation.

Unfortunately, many ML projects fail at various stages without ever reaching production. However, with the correct approach and a mix of technical and business expertise, such as that provided by Fractal’s data science team, it is possible to avoid or quickly resolve many of these common pitfalls. Fractal can help organizations deploy more ML models to production and achieve a faster time to value for ML projects with the tools, practices, and processes of MLOps.

Starting with the business objective

Fractal’s proven MLOps methodology helps streamline and standardize each stage of the ML lifecycle from model development to operationalization. It allows collaboration between technical and non-technical users and empowers everyone to participate in the development process actively.

We have helped many organizations leverage MLOps, allowing them to overcome their challenges. It includes a process for streamlining model training, packaging, validating, deployment, and monitoring to help ensure ML projects run consistently from end to end.

Our successful 5-stage model

Model development diagram

Train: We create and train an initial model based on available data, business requirements, and desired outcomes.
Package: Once the model is trained, we package up the model to make it easy to test, iterate, and deploy at scale.
Validate: Later, we help validate models by measuring candidate models against predefined KPIs, deployment testing, and testing application integrations.
Deploy: On validating, we deploy models by identifying the deployment target, planning the deployment, and then deploying the models to their production environment. We ensure that the services are implemented to support scalability, data pipelines are automated, and a model selection strategy is implemented.
Monitor: Finally, we monitor models to track behavior, continuously validate KPIs, measure accuracy and response times, watch for drift in model performance, and more.

Google Cloud services for ML model deployment

We can help teams successfully deploy and integrate more ML models into production and achieve a faster time to value for ML projects with more efficient model training using Google Cloud services such as:

Google Cloud Storage: It enables organizations to store, access, and maintain data so that they do not need to own and operate their own data centers, moving expenses from a capital expenditure model to an operational expenditure.
Cloud Functions: It provides a simple way to run code responding to events with minimal configuration and maintenance. Cloud Functions are event-driven, meaning they can be triggered by changes in data, new messages, and user interactions.
Big Query: A fully managed enterprise data warehouse helps you manage and analyze your data with built-in features like machine learning, geospatial analysis, and business intelligence.
Kubernetes engine: A solution to help organizations achieve zero ops. Kubernetes is an open-source container orchestration system for automating software deployment, scaling, and management.
Data Proc: It is a fully managed and highly scalable service for running Apache Hadoop, Apache Spark, and 30+ open-source tools and frameworks.
Vertex AI: It is a machine learning platform that lets you train and deploy ML models and AI applications and customize large language models (LLMs) for use in your AI-powered applications frameworks.

Leveraging both business and technical expertise

MLDEVOps diagram

Our ML model lifecycle unifies data collection, pre-processing, model training, evaluation, deployment, and retraining to a single process that teams maintain. It allows businesses to quickly tackle obstacles faced by the data scientists and IT operations team while providing a mix of technical and business expertise.

How Fractal’s methodology benefits organizations

Eliminates guesswork
Supports consistency
Enables continuous packaging, validation, and deployment of models to production
Rapid time to value
Accelerate time-to-value and time-to-deployment
Efficiently manage data error and model performance
Increase model scalability during training and serving

Conclusion

As we continue through 2023, the MLOps market is surging rapidly. As ML applications become a key component for maintaining a competitive advantage, businesses realize they need a systematic and reproducible way to implement ML models. According to the analyst firm Cognilytica, MLOps is expected to be a $4 billion market by 2025. Fractal has deep expertise in MLOps and can help deliver solutions for unique business challenges across virtually all industries and sectors.

Ready to begin leveraging MLOps in your organization? Contact us to get started.

3 key benefits of DevOps

Technology projects require maintenance throughout the entire lifecycle, from development to deployment and monitoring. Maintaining a project from version to version can become a manual and strenuous process. Developers must take special considerations at each stage to ensure smooth rollouts. Failure to do so can result in extended-release planning cycles ensuring the software is ready for use by end users.

Development and IT Operations teams may end up spending unnecessary cycles supporting and fixing buggy software. And even worse, failed software releases can impact a company’s financial performance through operations inefficiencies, lost sales, and customer attrition. Failure to maintain working software can impact SLAs, regulatory compliance for some industries, resulting in fines or legal action.

Successful organizations have adapted and created a set of best practices for governing projects, called DevOps.

DevOps illustration

What is DevOps?

DevOps aims to create a common culture that brings together the people, processes, and technology to deliver value (i.e., working software) to end-users.

It has also come up with procedures for automating many manual maintenance steps to reduce the time it takes to develop, test, and deploy software. Many companies are rushing to implement DevOps to avoid the high costs associated with manually maintaining projects.

Benefits of DevOps on Google Cloud

If you’re asking this question, keep reading. Outlined below are three key benefits of implementing DevOps in your organization.

Benefits of DevOps

1. Improved quality

Standardized tools and processes (i.e., Google Cloud DevOps and Agile Project Management) help keep quality consistent across projects and releases
Increased DevOps flow on Google Cloud helps improve software quality, delivery, and operations which leads to maintained security and faster deployment from the start
Quality control implemented through source control branching, code reviews, environment management, release management, etc.
Reduced fire drills and break-fix measures because of following DevOps best practices

2. Reduced effort

Fewer manual processes and interventions through improved automation
Improved software stability leads to faster deployment with minimum manual intervention
Lower effort to support/maintain because solutions have gone through the appropriate governance and management processes
Leverage templates for common infrastructure builds/configurations to accelerate new projects going forward

3. Increased collaboration

Agile project management structure encourages frequent collaboration among team
Google Cloud provides robust CI/CD capabilities, allowing teams to automate the entire software delivery pipeline. By integrating development, testing, and deployment processes, DevOps teams can collaborate more effectively and achieve faster and more reliable software releases
Improved communication channels enable the team to identify, track, and manage risks as they arise
Clear direction and prioritization through collaboration between stakeholders, developers, and end-users

Hopefully, this helps you better understand some of the benefits that implementing DevOps using Google Cloud can bring to your business.

Implementing DevOps is a journey and is not as easy as installing a package, flipping a switch, or buying new technology. Fractal specializes in helping our customers through the process of implementing DevOps, no matter what their current maturity level is. Fractal can provide strategic planning, technology assessments, proof of concepts, as well as technical resources to get you started on your DevOps journey.

Interested in learning more about how Fractal can help you implement DevOps? Please contact us for additional information from our experts.

Three pillars of the retail industry: Replenishment, allocation, and transportation 

The retail industry is one of the most dynamic and fast-paced sectors, comprised of various stakeholders engaged in selling finished products to end-user consumers. In 2022, the U.S. retail sector was estimated at more than seven trillion dollars. The sector is projected to continue to grow, and by 2026 U.S. retail sales are expected to reach approximately USD 7.9 trillion. With the increased demand for consumer goods in different sectors and the ever-increasing choices of products at low costs, investments in the retail sector have also grown over the past few years.

As there is always an element of change in this industry, new challenges are encountered daily. Take product stockouts, for example. Suppose a customer walks into a grocery store to purchase items of their favorite brand but discovers the product is unavailable. Frustrated by this, the customer chooses to either buy another brand or postpone the purchase; both scenarios are unfavorable to the business. The brand image and sales of the product are damaged because of this out-of-stock issue. The out-of-stock situation occurs when the inventory of a particular product is exhausted; this causes a problem for suppliers and retailers.

Multiple reasons could cause the product stockout, such as inaccurate inventory data, lack of demand forecasting, or an unseasonal spike in purchasing. Many of these underlying causes of stockouts can be avoided if the business implements adequate processes to be carried out every month.

To avoid situations like the stockout example above, retail companies need to develop a methodology for streamlining the following operations:

3 pillars of retail industry

Replenishment
Allocation
Transportation

These three operations create the three pillars of the retail industry that help monitor real-time insight into customer behavior and understand their buying patterns, hence strengthening the retail foundation.

1. Replenishment

Replenishment refers to a situation where the amount of stock left in the store is counted so that the right products are available in an optimal quantity. It is considered an essential aspect of inventory management as it ensures that the right products are being reordered to meet the customer demand.

In operational terms, the efficiency of store replenishment has a significant impact on profitability. The effectiveness and accuracy of store ordering affect sales through shelf availability and storage, handling, and wastage costs in stores and other parts of the supply chain. By optimizing demand forecasting, inventory management, and setting order cycles and order quantities by making them more systematic, the gains obtained are significant, often amounting to savings of several percent of total turnover.

For companies that must manage many SKUs, one of the most effective ways of making store replenishment more accurate, efficient, and cost-effective is by using a replenishment system specifically tailored to business operations. When many different products need to be managed, manual ordering is highly labor-intensive and expensive; this results in companies using the replenishment system.

An efficient replenishment system reduces process costs, improves inventory turnover, and provides higher service levels. The system constantly monitors the stock, sales, and demand while considering the forecast changes in demand and adjusting the replenishment orders. The company can control its inventory to ensure long-term profitability by recognizing the sales frequency, value, or profit margin. The replenishment system calculates the safety stock level for each SKU separately and sets them to meet the service level targets with efficiency, considering the predictability of demand.

2. Allocation

In allocation, the new stock of products is distributed to individual store units to maximize the product’s sales and prevent any stock-out situation in the future. This process enables assigning supplies to support the organization’s strategic goals. Having sufficient stock levels is an essential component for any retail business; with the changing consumer habits, it becomes crucial for the stock to be available in the right place at the right time.

3. Transportation

Transportation plays a significant role in delivering the right stock of products at the right point of delivery. It connects the business to its supply chain partners and influences customer satisfaction with the organization. With the ever-changing customer preferences and as their expectations continue to evolve, the transportation industry is undergoing a dramatic transformation to meet these demands.

Today, data plays a vital role in shaping the industry’s progress amidst tough competition. Due to the maturation of automation technologies, AI will help the transportation industry to manage drivers and fleet managers. By employing the techniques of AI, fleet and truck adjustments will offer data in real-time, eventually improving the industry’s standard. The safety and retention of the drivers will also increase from these newly acquired standards, and with enhanced access to data, there will be transparent communication between drivers and carriers.

These three pillars of retail can be strengthened by adopting in-house solutions and capabilities like StockView for retail, Edge video analytics, and Dynamics 365 customer insights.

How could these solutions help the retail industry?

StockView for retail helps retailers reduce lost sales and improve customer experience by automatically detecting out of stock items on shelves. It uses computer vision technology running at the edge to detect gaps on store shelves automatically. It also provides retailers with powerful insights and analytics into stock-out activities at both single-store and multi-store levels.

Powered by Microsoft Azure Stack Edge, it offers a scalable, flexible, and cost-effective solution that brings the power of the Azure cloud platform down to the individual store, eliminating the need for costly and unreliable data transfers while offering a predictable and consistent TCO (Total Cost of Ownership).

Edge Video Analytics solutions for retail typically leverage Azure Stack Edge devices, IoT devices, vision AI (Artificial Intelligence), and other technologies in conjunction with in-store cameras or other video sources.

Insights gained from Edge Video Analytics can allow retailers to swiftly react to in-store customer behavior and product stock-outs while improving security and reducing shrinkage.

Also, organizations can track store foot traffic, automatically notify employees to open more checkouts due to lengthy queues, and automatically detect product stock-outs. These insights can be used to help improve demand forecasting, optimize supply chains, and detect variances in population preferences down to the individual store.

Dynamics 365 customer insights give you access to Microsoft’s extensive library of pre-built data connectors to help businesses gain faster time to value. Plus, you can power up your customer profiles with AI by leveraging Azure Machine Learning and Azure Cognitive Services, and Fractal’s library of custom AI models. So, we leverage this solution to maximize the potential of customer data.

All these solutions and capabilities help understand the customer motivations, preferences, and desires to meet their demands and increase sales effectively, strengthening the retail industry’s pillars.

Conclusion

To meet these growing customer expectations, retailers should prioritize collecting customer data and analyzing it to support business decisions throughout their value chain. The inventory stocking patterns and shipping routes will shift in relation to patterns informed by this data. Retailers should make a concentrated effort to leverage the data while making critical business decisions, and to remain efficient; they must remain flexible and transform their operations as they capture more insights from their data.

Find how Fractal helped a global CPG company to operationalize an analytics engine designed and provide store recommendations for maximizing investments in their Azure environment. Read the full case study here: https://fractal.ai/casestudies/store-level-insights/

Leveraging generative AI to boost the productivity of field technicians and engineers

GenAI for field technicians and engineers

Field technicians and engineers are the backbone of various industries, including manufacturing, energy, telecom, and more. They play a crucial role in installing, repairing, maintaining, inspecting, and troubleshooting equipment and systems — often in challenging environments.

One of the most significant pain points they encounter is the difficulty of accessing essential information from thick manuals and field service guides while working in the field. Advances in Generative AI (or GenAI), however, can revolutionize the way field technicians and engineers access information and streamline their daily tasks.

Common challenges faced by field technicians

Field technicians and engineers encounter a lot of challenges daily in their line of work. These challenges vary depending on the industry and environment they operate in, but they include:

Complex tasks: Field technicians often deal with intricate machinery and systems that require specialized knowledge and expertise to operate, maintain, and repair.
Constantly changing environments: Whether it’s working in remote locations, extreme weather conditions, or high-pressure situations, field technicians must adapt to diverse and unpredictable work environments.
Information overload: Technicians are frequently required to consult technical manuals, documentation, and safety guidelines to perform their tasks accurately. The sheer volume of information can be overwhelming.
Time constraints: Efficiency is crucial, and technicians must perform tasks swiftly without compromising safety or quality.
Customer expectations: Meeting customer expectations for prompt service and problem resolution is paramount for customer satisfaction and loyalty.
Wide range of field technicians’ experience levels: While experienced technicians may be able to quickly find the relevant information, more junior ones can struggle to effectively find it.

How GenAI can help

Recent advances in GenAI offer natural language understanding and generation capabilities, making it an ideal solution for field technicians and engineers. It enables them to use voice and text to ask questions in plain English and receive well-articulated answers grounded in products, systems, and processes documentation.

GenAI for field technicians and engineers

Here are some examples of questions that GenAI could reply to:

How do I replace the filter of machine model XYZ?
What is the safety process for operation X?
What are the potential causes of error code 1234?
How do I calibrate sensor model XYZ?

Benefits of using GenAI for field service work

GenAI can help both transform the worker experience and deliver immense value for the customer. Here we discuss a few benefits Gen AI delivers:

benefits of using GenAI

Efficiency: GenAI reduces the need to search and browse through lengthy and complex manuals on small screens, allowing technicians to find information more quickly and efficiently.
Accuracy: GenAI provides instant and relevant answers based on the latest approved documentation, minimizing the risk of errors and accidents.
Convenience: With hands-free and voice-based interaction, technicians can access information without the need to type or tap on mobile devices, enhancing convenience especially in scenarios where typing is not the best option: gloves, greasy hands, etc.

Customer satisfaction: Faster and more reliable service delivery translates to improved customer satisfaction, leading to increased trust and loyalty.

FractalGPT for field service work

FractalGPT is a Google Cloud Vertex AI-GenAI so called “LLM”-powered AI chatbot solution that enables field service technicians to quickly find accurate information from company products and process documentation using a natural language interface. It helps enhance the productivity and efficiency of field technicians and engineers.

Some key features and benefits are:

key features and benefits

Device compatibility: FractalGPT is accessible on any device, whether it’s a phone, tablet, or laptop, via its responsive web app running on your private cloud tenant.
Customization: Tailor FractalGPT to your company’s machines, processes, and policies, to ensure it meets your specific needs.
Content breath: FractalGPT can access and analyze any type of documentation, including manuals, guides, datasheets, and more.
Contextual answers: It generates precise answers that are tailored to the context and user preferences, improving the user experience.
Complex query handling: FractalGPT can handle complex, multi-step queries that require reasoning and inference, ensuring that technicians get the information they need to complete tasks successfully.

Get started with FractalGPT

In today’s fast-paced world, field technicians and engineers need efficient solutions to overcome the challenges they face daily. FractalGPT offers a transformative approach to addressing these challenges.

It is designed to be secure, scalable, and easy to use, making it ideal for businesses of all sizes and across industries. With FractalGPT, your teams can improve productivity while ensuring your data remains secure.

Leveraging our partnership with Google Cloud and decades of AI expertise, Fractal can swiftly and efficiently assist you in implementing your Generative AI solution using FractalGPT.

Take the first step in improving your field service operations by deploying your custom-branded FractalGPT in a week.

Consumer Packaged Goods>

Retail>

Industrials, Manufacturing, and Energy>

Healthcare & Life Sciences>

Financial Services>

Technology, Media, Telecom>

Insurance>

Capabilities

AI and Data Engineering>

Generative AI>

Finance Analytics>

Dimension>

Supply Chain>

Responsible AI>

Quantum Computing>

IME>

ESG>

FAA>

FinalMile>

Customer Experience>

MLOps>

Solutions

Trial Run>

Customer Genomics>

Image & Video Analytics>

Cognitive automation>

Conversational AI>

AIDE>

AI @ Scale>

Caliper>

Consumer Hub>

Concordia>

Products

Asper.ai>

Crux Intelligence>

Eugenie.ai>

Flyfish.ai>

Kalaido.ai>

MarshallGoldsmith.ai>

Qure.ai>

Senseforth.ai>

Overview>

Webinars>

Client Advisory Board (CAB)>

AI Series>

Life at Fractal>

Job Openings at Fractal.ai>

Job Openings at Fractal Alpha>

ReBoot>

CEO message on COVID-19>

Our Values>

Leadership>

Newsroom>

Partnerships and Alliances>

Corporate Social Responsibility (CSR)>

Annual Report>

Key benefits of AI in banking:

More in Related

From Storage to Insights: Streamlining Your Data Pipeline with Redshift and S3

Fractal Achieves the AWS Generative AI Competency

A comprehensive guide to adopting Microsoft Fabric within your organization

Step 1: Set up communication between apps

Step 2: Create a Redshift Cluster Using Boto3

Troubleshooting:

Step 3: Set Up an Inbound Connection

Step 4: Transfer Data from S3 to Redshift

Step 5: Delete the Cluster (Optional)

Need Help with Implementation?

More in Related

Unlocking the Future of Banking with AI

Fractal Achieves the AWS Generative AI Competency

A comprehensive guide to adopting Microsoft Fabric within your organization

What’s Next?

About Fractal

More in Related

Unlocking the Future of Banking with AI

From Storage to Insights: Streamlining Your Data Pipeline with Redshift and S3

A comprehensive guide to adopting Microsoft Fabric within your organization

More in Related

Unlocking the Future of Banking with AI