Open-Source Fund Data Chatbot

I already maintain a well-structured MySQL database that stores exhaustive mutual-fund research—historical returns, risk metrics, holdings breakdowns, ratings, fees, and manager information. What I now need is a lightweight AI layer that lets an end-user type a plain-language question such as “Which large-cap funds beat the benchmark on a 3-year risk-adjusted basis?” and immediately receive:

• a succinct narrative answer with the key numbers
• an optional drill-down table (and basic chart) that reveals performance, risk, and allocation metrics pulled straight from the DB
• a dynamic list of matching funds that can be further filtered inside the UI

Key requirements

— The system must rely solely on open-source models or other non-metered solutions; no paid-token APIs are acceptable.
— Natural-language understanding should comfortably cover questions about performance, risk, holdings, sector/asset allocation, ratings, fees, and manager details.
— All answers must be generated on-the-fly from the live MySQL data, not from a static snapshot.
— Results have to remain fully traceable: include a quick way for users (and me) to see which database fields fed each figure.
— A simple web front end is enough; React, Vue, or even a clean Flask template are fine as long as the UX stays fast and intuitive.
— You will hand over clean, well-commented code plus a short deployment guide so I can reproduce the setup on my own server.

Acceptance criteria

1. Typing a valid English query returns the correct funds and metrics in under two seconds on a sample data slice of 100,000 records.
2. Switching the data source (e.g., moving from staging to production) requires only a connection-string change.
3. The entire stack runs locally without external calls once the model is downloaded.
4. Code quality passes a brief peer review for readability and security (SQL injection, prompt injection, etc.).

If you have hands-on experience building retrieval-augmented chat or Q&A systems over large relational datasets, let’s talk. I am ready to share a sample dump and walk through the schema so you can map the fields to your NLP pipeline right away.