A safe, schema-aware Natural Language → SQL application that allows users to query a database using plain English. The system generates SQL using an LLM, validates it for safety, and executes it in a controlled, read-only manner.
This application allows users to ask questions about a database in natural language and receive query results without writing SQL manually.
Example flow:
User question → LLM generates SQL → Validator checks safety → Executor runs query → Results displayed
- 🤖 Natural Language to SQL
- 🧠 Schema-Aware Prompting (LLM sees actual DB schema)
- 🛡️ Read-Only SQL Enforcement (SELECT-only execution)
- 📊 Database Schema Viewer
- 💡 Sample Questions for Quick Start
- 🎨 Simple, Clean UI built with Streamlit
The application is structured into clear, single-responsibility modules:
- UI (app.py) – User interaction and result display
- Service Layer (service.py) – Orchestrates the NL → SQL pipeline
- LLM (llm.py) – Converts natural language to SQL
- Validator (validator.py) – Enforces SQL safety rules
- Executor (executor.py) – Executes validated queries
- Schema (schema.py) – Extracts database metadata
- Config (config.py) – Centralized configuration
This separation keeps the system debuggable, safe, and maintainable.
Multiple layers prevent unsafe execution:
- Only SELECT queries are allowed
- Dangerous keywords (DROP, ALTER, PRAGMA, etc.) are blocked
- Multiple SQL statements are disallowed
- Executor performs a final safety check before execution
- Errors are handled gracefully and surfaced to the user
- Fixed sample database (SQLite Chinook DB)
- Read-only queries only
- Complex or ambiguous questions may fail
- SQLite-specific schema introspection
- Not optimized for large datasets
- Python
- Streamlit
- SQLite
- Groq API
- Pandas
- LLaMA 3.3 70B
Built as a practical, production-minded AI engineering project 🚀