QUERYGEN

A natural language to SQL converter using GPT-4o-mini and semantic search. This project provides a web interface where users can input queries in plain English and receive generated SQL queries with their results.

---

🚀 Overview of Architecture

The system is implemented using a modular microservice architecture comprising the following components:

• Client (Next.js):
  Provides a user-friendly web interface to input queries in natural language and display results.

• Server (Node.js):
  Handles the generation of SQL queries using the GPT-4o-mini model, manages prompt engineering, and coordinates communication between services.

• Python Server (Flask):
  Computes the semantic similarity between the user query and database rows using the all-MiniLM-L6-v2 embedding model to select relevant context for SQL generation.

This architecture enables clear separation of concerns, scalability, and ease of maintenance.

---

🛠️ Running the Application

Follow these steps to set up and run the QUERYGEN application on your local machine.

Prerequisites

• Docker
• OpenAI API key

---

1. Clone the Repository

git clone https://github.com/suyashvsingh/QUERYGEN

---

2. Create .env file

Inside the QUERYGEN directory, create a .env file with the following content:

OPENAI_API_KEY=your_openai_api_key_here
MODEL=gpt-4o-mini

Replace your_openai_api_key_here with your actual OpenAI API key.

---

3. Run docker-compose.yaml file

cd QUERYGEN
docker compose up

---

💡 Usage

After completing the setup, open your browser and navigate to http://localhost:3000.
You can now input natural language queries and view the generated SQL queries and their results.

---

📁 Project Structure

QUERYGEN/
├── client/      # Next.js frontend
├── server/      # Node.js backend (SQL generation & coordination)
├── python/      # Python Flask server (semantic similarity)
├── .env         # Environment variables
├── docker-compose.yaml  # Docker Compose configuration
└── README.md

---

📊 Evaluation

The QUERYGEN system was tested using the Spider dataset, a benchmark for text-to-SQL tasks. The results across different difficulty levels are as follows:

Difficulty Level	Execution Accuracy (EX)
Easy	90.0%
Medium	73.0%
Hard	69.8%
Extra	65.0%
Overall	75.2%

Interpretation:

QUERYGEN achieves strong performance, especially on easy and medium queries, and maintains robust accuracy even on complex and extra-hard tasks.

---

🤝 Contributing

Contributions, issues, and feature requests are welcome!
Feel free to open an issue or submit a pull request.

---

📄 License

This project is licensed under the MIT License.