🤖 Natural Language to SQL Agent using IBM Watsonx.ai and LangChain

This project allows users to interact with a SQL database using natural language. It uses IBM Watsonx's Granite-3-2-8B Instruct LLM integrated with LangChain to dynamically generate SQL queries and fetch results from a MySQL database.

🔍 Ask questions like:

• "What is the total revenue by genre?"
• "List all albums by 'AC/DC'"
• "How many customers are from Canada?"

🧠 Key Features

• ✅ Converts plain English to SQL queries
• ✅ Uses IBM Watsonx's LLM (Granite 3-2-8B Instruct) via WatsonxLLM
• ✅ Executes queries securely on a MySQL database
• ✅ Handles SQL parsing errors and ensures verbose output
• ✅ Command-line interface with --prompt argument

🛠️ Technologies Used

• 🧠 IBM Watsonx.ai Granite-3-2-8B Instruct
• 🧱 LangChain
• 🐬 MySQL with Chinook Sample Database
• 🐍 Python
• 🔐 argparse for command-line usage

📁 Project Structure

project/
│
├── llm_agent.py        # Test script to invoke LLM only
├── sql_agent.py        # Main agent script that handles user queries
├── README.md           # Project documentation

⚙️ Setup Instructions

1. Clone the Repository

git clone https://github.com/aaradhya466/NLQ-to-SQL-Agent-IBM-Watsonx.git
cd NLQ-to-SQL-Agent-IBM-Watsonx

2. Install Required Packages

pip install ibm-watsonx-ai langchain mysql-connector-python sqlalchemy

💡 You may also need:
pip install argparse (in case it's not pre-installed)

3. Setup IBM Watsonx Credentials

Ensure you have access to IBM Watsonx.ai and replace the following inside sql_agent.py and llm_agent.py:

credentials = {
    "url": "https://us-south.ml.cloud.ibm.com"
}
project_id = "skills-network"

✅ You may also need to include apikey if required for your Watsonx instance.

4. Update MySQL Configuration

Replace the following lines with your own MySQL credentials in sql_agent.py:

mysql_username = 'your_username'
mysql_password = 'your_password'
mysql_host = 'your_host_or_ip'
mysql_port = '3306'
database_name = 'Chinook'

Ensure the MySQL server is running and the Chinook sample DB is imported.

---

🚀 How to Run

🔹 Option 1: Invoke LLM Only

python llm_agent.py

🔹 Option 2: Query Database Using Natural Language

python sql_agent.py --prompt "Show top 5 customers by total purchase"

---

🧪 Sample Prompts to Try

• "List all tracks longer than 5 minutes"
• "Find total sales by country"
• "Get names of all customers in France"
• "Show the names of employees and their job titles"

---

📌 Notes

• The project uses ZERO_SHOT_REACT_DESCRIPTION agent type from LangChain for SQL reasoning.
• Agent handles parsing errors and logs verbose output for debugging.
• The Chinook database must be running on the specified host and port.

---

👤 Author

Aaradhya
2nd Year BCA Student | Specialization in AI & Data Science
📧 Email: aaradhya8200@gmail.com
🌐 GitHub: @aaradhya466

---

🎯 Future Improvements

• Add frontend interface to interact with the agent
• Integrate with other SQL dialects (e.g., PostgreSQL, SQLite)
• Secure credential management via .env or vaults