🤖 LLM-Powered SQL Query Agent (FastAPI + Groq + SQLite)

🧠 Introduction

The LLM-Powered SQL Query Agent is an AI-driven system that enables users to query structured databases using natural language. Instead of manually writing SQL queries, users can ask questions in plain English, and the system automatically converts them into optimized SQL statements using a Large Language Model (LLaMA 3 via Groq API).

This project demonstrates the integration of:

• Generative AI (LLMs)
• Backend API development (FastAPI)
• Secure SQL query validation
• Database management (SQLite)

The system ensures that only safe and read-only queries are executed, preventing data corruption and SQL injection attacks.

---

❗ Problem Statement

Traditional database querying requires:

• Knowledge of SQL syntax
• Understanding of table relationships
• Ability to write joins, filters, and aggregations

This creates a barrier for non-technical users.

This system solves this by:

• Translating natural language → SQL
• Ensuring secure query execution
• Returning structured and readable results

---

🎯 Objectives

• Convert natural language into SQL queries
• Automatically extract database schema
• Validate SQL queries for security
• Execute safe queries on SQLite database
• Return structured JSON responses
• Prevent SQL injection and destructive operations

---

🏗️ System Architecture

User → FastAPI (API Layer) → Agent Pipeline → Groq LLM (LLaMA 3.3-70B) → SQL Validator → SQLite Database → JSON Response

---

⚙️ Technology Stack

• Backend: FastAPI
• LLM: LLaMA 3.3-70B (Groq API)
• Database: SQLite (Chinook Dataset)
• Language: Python 3.10+
• Environment: python-dotenv
• Server: Uvicorn

---

🔄 Working Methodology

🧩 Schema Extraction

The system extracts database structure using:

• Table names
• Column names
• Schema formatting for LLM context

👉 Ensures the model understands database structure before generating SQL.

---

🤖 SQL Generation (LLM)

Function: generate_sql(question, schema)

• User question + schema sent to LLM
• Model: llama-3.3-70b-versatile (Groq)
• Output: optimized SQL query

---

🔐 SQL Validation (Security Layer)

Function: validate_sql(query)

Security checks:

✔ Allows only SELECT queries
❌ Blocks:

• DROP
• DELETE
• INSERT
• UPDATE
• ALTER
• TRUNCATE

👉 Prevents SQL injection and destructive operations.

---

💾 SQL Execution

Function: execute_sql(query)

• Executes query on SQLite database
• Fetches results
• Converts output to JSON format

---

🌐 API Documentation

🔹 Endpoint

GET /ask

http://127.0.0.1:8000/ask?question=How many tracks are in the Rock genre?

🔹 Example Response

{
  "response": {
    "question": "How many tracks are in the Rock genre?",
    "sql_query": "SELECT COUNT(*) FROM Track t JOIN Genre g ON t.GenreId = g.GenreId WHERE g.Name = 'Rock';",
    "answer": "COUNT(*): 1297"
  }
}

🚀 Key Highlights
Natural Language → SQL conversion using LLM
Secure SQL validation layer
Read-only database enforcement
Real-time query execution
API-based architecture using FastAPI
🔐 Security Features
Only SELECT queries allowed
Blocks destructive SQL operations
Prevents SQL injection attacks
Validates queries before execution
📈 Future Improvements
Support for multiple databases (MySQL/PostgreSQL)
Advanced query optimization
Role-based access control (RBAC)
Web-based UI dashboard
Multi-table reasoning improvements
🏆 Project Impact

This project demonstrates:

Real-world application of Generative AI
Secure backend system design
Database interaction using LLMs
Production-style API architecture

It highlights skills in AI + Backend + Security + System Design, making it highly relevant for research and abroad scholarship applications.

👩‍💻 Author

Charuhasini
AI & Data Science Student
GitHub: https://github.com/Charuhasini30

---

### 🔹 Example Request