Unlocking Building Information Models (BIM) Insights with Fractal’s GenAI Copilot powered by AWS Bedrock
Overview
A global leader in engineering and design software partnered with Fractal to revolutionize how its users interact with Building Information Models (BIMs). Traditionally, engineers and architects spent hours navigating technical drawings and documents to locate specific details, slowing decision-making and delaying project delivery.
Fractal developed an Intelligent BIM Navigator, powered by Amazon Bedrock, that enables users to query BIMs using natural language. By fine-tuning large language models and combining them with a custom Retrieval-Augmented Generation (RAG) framework, Fractal delivered instant, domain-aware responses from deeply technical project data.
The result: a significant reduction in manual search time, faster insights, and improved efficiency across planning and engineering workflows.
Challenge
BIMs are rich in data but difficult to navigate. Engineers, architects, and planners often spend hours digging through complex design files to find specific information, such as materials, layouts, or compliance details.
This time-consuming process delays decisions and diverts focus from higher-value work. The client needed a faster, smarter way to interact with BIM data, ideally by allowing users to ask questions in natural language and receive precise answers instantly.
Solution
Fractal developed Copilot on AWS to simplify how professionals interact with complex design data. This solution blends the power of Large Language Models (LLMs) with a robust, scalable architecture, delivering natural language access to highly technical BIM insights.
At the heart of the solution are fine-tuned Mistral 7B models hosted on Amazon Bedrock, trained to understand user intent and convert everyday questions into Graph Query Language (GQL). To strengthen accuracy, Claude 3.5 Sonnet was used to generate synthetic training data and craft human-like responses that align with the technical language of the AEC domain.
Understanding that each query needs to reflect deep domain knowledge, Fractal implemented a custom Retrieval-Augmented Generation (RAG) pipeline. Technical documents, including design specifications and compliance codes, were stored in Amazon S3, processed into searchable chunks, and embedded via Amazon OpenSearch Service. This ensured that responses weren’t just linguistically accurate, but also contextually precise.
To bring this intelligence into real-world usage at scale, the solution was deployed using Amazon SageMaker for hosting model endpoints and integrated securely with AWS Lambda and API Gateway, enabling seamless access for users across different applications and roles.
This fully integrated approach turned what was once a time-intensive, manual task into an instant, AI-powered experience.
Impact delivered
The Fractal’s GenAI Copilot delivered both measurable performance gains and strategic transformation:
- Significant reduction in manual search efforts
Professionals now retrieve project-critical insights in seconds, not hours.
- Faster decision-making
Real-time interaction with BIMs accelerated key stages in design and planning.
- High-accuracy in generated GQL queries
The model delivered precise, trustworthy outputs across diverse project types and use cases.
- Enhanced user experience and adoption at scale
Intuitive natural language interaction drove faster adoption across roles and disciplines.