Talk to Your Data — Natural-Language Analytics
A live, gated demo: load a CSV (or a built-in sample) and ask questions in plain English. The AI writes SQL, runs it against an in-memory SQLite engine, and returns a results table plus an auto-generated chart. LLM-agnostic; this instance uses Gemini.
Overview
Talk to Your Data is a working proof-of-value for natural-language analytics. Authenticated demo users load a CSV (or pick a built-in sample), then ask questions conversationally — "total revenue by region", "top 5 by salary", "revenue trend by quarter". The app translates the question into a single read-only SQL query, executes it against an in-memory SQLite engine over the uploaded data, and returns both a results table and an auto-generated chart, alongside the exact SQL it ran for full transparency. Only read-only SELECT queries are ever executed. The architecture is LLM-agnostic — the same flow runs on a local model via Ollama (e.g. Llama 3) or any cloud LLM; this deployment uses Google Gemini. Data is processed in-memory and expires automatically.
Problem Solved
Business users wait on analysts for every ad-hoc question, and analysts spend time writing repetitive SQL. This demo shows how an LLM, constrained to safe read-only queries and grounded in the actual table schema, lets anyone interrogate a dataset in plain language — while still exposing the generated SQL for trust and review.
Capabilities
- CSV upload or built-in sample datasets
- Natural-language question → SQL translation
- Real query execution on an in-memory SQLite engine
- Read-only guardrail (SELECT-only, single statement)
- Results table plus auto-generated bar/line charts
- Transparent display of the generated SQL
- Pluggable LLM backend — Ollama / Llama (local) or cloud LLM
- Gated access, in-memory processing, rate limiting