CodeCraft

CodeCraft is an AI-powered iterative code refinement tool that utilizes Large Language Models (LLMs) to generate, test, and refine code until it meets the expected output. It follows a structured feedback loop incorporating Chain of Thought prompting techniques for enhanced reasoning and explainability.

Features

• Automated Code Generation: Generates code based on a given programming question, example input, and language.
• Iterative Refinement: Uses test cases and feedback loops to debug and enhance code quality.
• Edge Case Handling: Automatically generates five test cases, including edge cases.
• Compiler Execution: Runs the generated code and verifies outputs against expected results.
• LLM Debugging Assistance: If the output doesn't match expectations, the model iteratively refines the code.
• Supports Multiple Programming Languages: Users can specify their preferred language for code generation.
• Groq API Integration: Users can input their Groq API key to interact with the application.

Deployment

Deployed URL: https://www.codecraftai.tech

Backend

The backend is built using FastAPI and utilizes PostgreSQL as its database.

Setup Instructions

1. Set up environment variables: Create a .env file in the root directory with the following variables:

   COMPILER_API_ENDPOINT=
   DATABASE_URL=
   CORS_ORIGINS=

   • COMPILER_API_ENDPOINT: URL of Judge0 API (either self-hosted or from RapidAPI).
   • DATABASE_URL: PostgreSQL database connection string.
   • CORS_ORIGINS: Allowed origins for CORS ie http://localhost:3000 for frontend

2. Set up a Python virtual environment:

   cd backend
   python -m venv venv
   source venv/bin/activate  # On Windows use `venv\Scripts\activate`
   pip install -r requirements.txt

3. Run database migrations:

   cd backend/src
   # edit alembic.ini from script_location = src/migrations to script_location = migrations
   alembic upgrade head

4. Start the server:

   cd backend/src
   uvicorn llama_agent:app --reload

Frontend

The frontend is built using Next.js.

Setup Instructions

1. Navigate to the frontend directory:

   cd frontend

2. Create an .env file: Copy .env.example and fill in the necessary values:

   NEXT_PUBLIC_API_URL=

3. Install dependencies and start the frontend:

   yarn install
   yarn dev

Workflow Overview

1. User Input:
   • Model selection
   • Programming question
   • Language
   • Example input
   • Number of iterations
2. Test Case Generation:
   • The model generates five test cases, including edge cases.
3. Code Generation:
   • Code is generated based on the input parameters.
4. Compilation and Execution:
   • The code is compiled and executed using the Judge0 API.
5. Verification:
   • The output is compared with expected results.
6. Feedback Loop:
   • If errors occur or outputs do not match, additional debugging information is fed back to the model for refinement.
7. Final Output:
   • Once the expected output is achieved, the final refined code is presented to the user.

Contributing

Contributions are welcome! Feel free to fork this repo and submit pull requests.

License

This project is licensed under the MIT License.