Agentic analytics over 100M+ rows

Naive RAG wouldn't cut it — so I built an agent that writes and runs its own queries.

From the RFP phase with the IDB: a chatbot over survey datasets in the hundreds of millions of rows. Knowing naive RAG would fail at that scale, I built an agent-orchestration system with a code-execution tool that queries the database and runs Python analysis on demand. Delivered, and praised by the bank's researchers alongside the document chatbot.