From Tribal Knowledge to Instant Answers: Building Reffy on Databricks

Databricks built Reffy, an internal RAG-based app enabling 1,800+ sales and marketing employees to discover and analyze 2,400+ customer references.

• •Customer stories from YouTube, LinkedIn, Medium, internal slides, and databricks.com are consolidated into a Bronze/Silver/Gold Delta Lake pipeline in Unity Catalog
• •Gemini 2.5 applies a 31-point scoring system to evaluate story quality and extract metadata like industry, products used, and quotes
• •A DSPy agent enables hybrid keyword and semantic search via Databricks Vector Search with a re-ranker for improved retrieval
• •The app uses a React frontend, FastAPI backend, and Lakebase for conversation persistence, deployed via Databricks Apps
• •In two months, 7,500+ queries were run; an AI/BI Dashboard monitors usage and identifies content gaps