Skip to content
Back to projects
Featured
BaxterDec 2023

Text2SQL for Finance

Led the foundation and production deployment of a Text2SQL assistant for the finance department, serving over 1,500 monthly users.

LLMFine-tuningMLOps
A finance question in plain English turned into SQL and a result table

Context: Baxter International (Healthcare)

Problem

The finance department needed answers from its data, but most users don't write SQL.

Approach

  • Foundation to production. I led the project from its foundation to production deployment for the finance department.
  • Text2SQL pipeline. Built with LlamaIndex and LangChain, turning natural-language questions into SQL over finance data.
  • Quality. Improved answers through prompt tuning and fine-tuning.
  • Open-source models. Managed the open-source models behind the assistant, served with vLLM.

A question goes through the Text2SQL pipeline to an open-source model, and the generated SQL runs against finance data

Impact

  • Serves 1,500+ monthly users in the finance department.

Stack

Python · LlamaIndex · LangChain · FastAPI · vLLM