Project
AI Data Assistant
Ask about US public data in plain English.
A chat assistant that answers questions about US population, income, housing, education, and demographics. You ask the way you would say it out loud, and it does the rest: works out the query, pulls the numbers from a cloud database, and replies in plain English.
How it works
- You ask a question in plain English.
- An LLM encodes your question into a precise database query.
- The query runs against a cloud SQL data warehouse holding the data.
- The results are turned back into a clear, written answer, streamed live.
your question (plain English) ↓ LLM encodes it into SQL ↓ runs on the cloud SQL warehouse ↓ rows become a written answer
Under the hood
- Natural language, encoded to SQL. Instead of you writing code, the LLM writes the query for you.
- Grounded in the data. The real columns are cryptic codes, so the app relies on the dataset's own descriptions, and pins the important metrics, so it picks the right column instead of guessing.
- Reliable by design. Several LLM providers with automatic failover keep it running, and guardrails keep it on topic and safe.
- Honest answers. It sanity-checks its own numbers and hedges when unsure, rather than sounding confident about a shaky figure.
Python
FastAPI
LLM encoding → SQL
Groq · Gemini · DeepSeek
cloud SQL warehouse
SSE streaming