Text-to-SQL LLM Copilot (FastAPI + SQLite)

Turn natural language questions into safe SQL over your data.
Built with FastAPI, SQLAlchemy, sqlglot guardrails, and an optional OpenAI LLM.

✨ Features

• Text → SQL: Rule-based baseline or LLM (OpenAI) with few-shots
• Safety: Read-only enforcement via sqlglot (blocks DROP/UPDATE/DELETE/DDL)
• Dynamic schema: LLM sees your actual SQLite tables/columns at runtime
• All-click VS Code workflow: Run, seed/load, and test via launch configs—no bash

🧩 Repo Structure

src/ api/ app.py # FastAPI app (/health, /query) core/ config.py # Settings (.env via python-dotenv) db/ seed_db.py # Small demo seed load_csvs.py # Load real CSVs -> SQLite nlp/ pipeline.py # Baseline & LLM SQL generators sql/ runner.py # Validate+run SQL safely (sqlglot) .vscode/launch.json # Click-to-run configs requirements.txt

🚀 Quickstart (VS Code only)

1. Create .env in project root:

OPENAI_API_KEY=sk-... # optional (only if using LLM) MODEL_PROVIDER=openai MODEL_NAME=gpt-4o-mini USE_LLM=true

2. Install deps (VS Code → Python: Manage Packages) or right-click requirements.txt → Install All.
3. (Optional) Load your CSVs

• Put customers.csv, products.csv, orders.csv, order_items.csv into data/real/
• Run Load real CSVs (module: src.db.load_csvs)

4. Start API: Start FastAPI (Uvicorn) → open http://localhost:8000/docs
5. Try POST /query:

{ "question": "revenue by category in 2024", "limit": 10 }