AI-Powered NL → SQL Analytics Platform

An intelligent data analytics system that allows users to query a MySQL database using natural language. The system converts user queries into SQL using a local LLM (Qwen via Ollama) and returns structured results.

---

🧠 Problem

Non-technical users struggle to write SQL queries.

👉 This system solves it by enabling:

Natural Language → SQL → Insights

---

⚙️ Tech Stack

• Frontend: Streamlit
• Backend: Python
• Database: MySQL
• LLM: Qwen2.5:3B (Ollama)

---

🏗️ Architecture

User → Streamlit → LLM (Qwen) → SQL → MySQL → Result

...

---

📂 Project Structure

ai-nl-to-sql-analytics-platform/
│
├── README.md       
├── requirements.txt
├── .gitignore
│
├── src/            
│   ├── app.py              # Streamlit UI
|   ├── backend.py          # Database logic, LLM integration, Helper functions
│   └── prompt.py           # Prompt engineering 
│                          
├── db/
│   ├── schema.sql      # CREATE TABLE statements
│   └── data/
│       ├── customers.csv
│       ├── orders.csv
│       ├── order_items.csv
│       ├── products.csv
│       ├── category.csv
│       └── departments.csv
|
└── images/

---

▶️ Setup Instructions

1. Clone repo

git clone <repo_url>
cd project

2. Install dependencies

pip install -r requirements.txt

3. Start Ollama

ollama serve

4. Run model

ollama run qwen2.5:3b

5. Run app

streamlit run app.py

---

🔥 Features

• Natural Language → SQL conversion
• Multi-table JOIN support
• Aggregation (SUM, COUNT, etc.)
• Prompt-engineered accuracy
• MySQL integration

---

Example Queries

Natural Language	Generated SQL
Show all customers	SELECT * FROM customers;
Total sales	SELECT SUM(total_amount) FROM orders;
Top 5 customers	GROUP BY + ORDER BY + LIMIT
Sales by department	Multi-table JOIN

---

Safety & Validation

• Blocks:

  • DROP
  • DELETE
  • UPDATE
  • TRUNCATE

• Ensures:

  • Read-only queries
  • Controlled execution

---

⚠️ Limitations

• LLM may generate incorrect queries in edge cases
• Depends on prompt quality
• Future RAG implimentation to reduce hallucination
• No dynamic schema support yet
• Limited scalability (local setup)

---

🚀 Future Improvements

• Dynamic schema detection
• RAG-based table retrieval
• Query validation & auto-correction
• Dashboard & charts
• Cloud deployment

---

🎯 Key Learnings

• Prompt engineering for structured tasks
• LLM integration with backend systems
• SQL query generation & optimization
• Building end-to-end AI systems

---

💡 Conclusion

This project demonstrates how AI can bridge the gap between users and data systems by enabling natural language querying over structured databases.

---

👨‍💻 Author

Mazhar Kakar