AI engineering case study
Supply Chain Copilot
Core stack
Multi-Agent AI · RAG · SQL Agents · Databricks · Genie MCP
Delivery approach
Situation
Enterprise questions crossed unstructured knowledge and structured data in Databricks Delta Tables. Analysts had to identify the right source, run the data query, and combine the results before responding, which slowed turnaround.
Task
Lead development of a multi-agent chatbot that could choose the right retrieval path for each question, query Delta-backed data through Genie MCP, and support more than 10,000 queries each month.
Action
Built separate RAG and SQL agents behind a hybrid query router. The router sends narrative questions to retrieval, structured questions to the SQL agent through Genie MCP, and mixed questions to both paths before combining the results into one response.
Result
The chatbot handles more than 10,000 queries each month and cut analyst turnaround time by 42%. Users can ask one question while the system selects the RAG, SQL, or hybrid path.
Technology stack
- Multi-Agent AI
- RAG
- SQL Agents
- Databricks
- Genie MCP