GenAI on the Lakehouse: a modern, governed foundation that turns Mazda’s existing service data into the backbone of a high‑impact GenAI assistant.

databricks

Mazda built a GenAI assistant for technical service hotline agents using Databricks Lakehouse, going from concept to working pilot in eight weeks with just two data scientists. The system uses RAG to connect an LLM with Mazda's corpus of service information documents, vehicle history, warranty data, recall status, and diagnostic codes. Unity Catalog provides governed access and hosts user-defined functions that filter vector search results to only documents applicable to a specific vehicle. MLflow 3's evaluation framework shifted the team from informal feedback loops to test-driven development with traceable experiment results. The architecture supports multilingual use without core changes, and a reusable governance pattern using service principals and role-based access control enables secure expansion to future GenAI applications.

From Legacy to Lakehouse: How Mazda Accelerated GenAI for Technical Service Operations

Starting Point: A Lean Team With Big Ambitions