Circuitry.ai builds RAG chatbots on Databricks to give manufacturers' customers decision intelligence

Overview

Circuitry.ai used Delta Lake, Unity Catalog, MLflow and Model Serving on Databricks to build customer-specific retrieval augmented generation chatbots, powered by Llama and DBRX, that let manufacturers of complex products like heavy equipment and automotive parts query proprietary product documentation instead of searching large PDF manuals. The RAG-powered Decision AIdvisor tool cut the time customers spend searching for information by 60-70%.

The challenge

Circuitry.ai's small technical team experienced delays developing RAG chatbots due to the complexity of managing and updating vast amounts of proprietary customer data, facing challenges applying metadata filters on retrievers, establishing internal checks for AI chatbots, handling knowledge base updates without disrupting internal RAG chains, and integrating multiple differently structured data sources.

The solution

Circuitry.ai used Delta Lake for incremental data updates to keep the knowledge base current, Unity Catalog for data governance and segregation of proprietary customer information, and MLflow and Databricks Model Serving to deploy and manage ML models, building custom RAG chatbots powered by Llama and DBRX with a feedback mechanism where users rate GenAI-generated responses for continuous improvement.

Reported business value

Customers of Circuitry.ai's decision intelligence software experienced a 60-70% reduction in the time spent searching for information, particularly speeding onboarding for new employees, and customers now receive answers in seconds rather than minutes instead of searching through large PDF files.