🧠 AutoCodeFix-AI: Transformer-Powered Automatic Code Repair

AutoCodeFix-AI is an advanced yet lightweight AI system that automatically detects, explains, and fixes code errors in Python and C-style languages using transformer-based models such as StarCoder2 and CodeT5+.

It provides:

• ⚙️ A FastAPI backend for scalable inference
• 🎛️ A Gradio web demo for live interaction and testing
• 📚 Notebooks for model finetuning and evaluation

---

🧩 Key Features

✅ Automated code error detection and correction
✅ Natural-language explanations of fixes
✅ Transformer-based fine-tuning (StarCoder2 / CodeT5+)
✅ Ready-to-use REST API (FastAPI)
✅ Interactive Hugging Face demo via Gradio
✅ Modular and lightweight (~40 MB project)

---

📁 Project Structure

autocodefix-ai/
│
├── README.md
├── requirements.txt
├── setup.py
├── LICENSE
│
├── data/
│   ├── raw/
│   ├── processed/
│   └── samples/
│
├── notebooks/
│   ├── 01_data_exploration.ipynb
│   ├── 02_preprocessing.ipynb
│   ├── 03_model_finetuning.ipynb
│   ├── 04_evaluation.ipynb
│   └── 05_inference_demo.ipynb
│
├── src/
│   ├── config/
│   │   ├── model_config.yaml
│   │   └── training_args.yaml
│   ├── data/
│   │   ├── data_loader.py
│   │   └── preprocessing.py
│   ├── model/
│   │   ├── model_builder.py
│   │   ├── train_model.py
│   │   ├── evaluate_model.py
│   │   └── inference.py
│   ├── utils/
│   │   ├── logger.py
│   │   ├── metrics.py
│   │   └── helper.py
│   └── app/
│       ├── main.py              # FastAPI backend
│       └── gradio_demo.py       # Local Gradio UI
│
├── deployment/
│   ├── Dockerfile
│   ├── api_service.py
│   ├── gradio_service.py
│   └── huggingface_spaces/
│       └── app.py
│
└── tests/
    ├── test_data.py
    ├── test_model.py
    ├── test_inference.py
    └── test_api.py

---

⚙️ Installation

git clone https://huggingface.co/spaces/hmnshudhmn24/autocodefix-ai
cd autocodefix-ai
pip install -r requirements.txt

---

🧠 Model Support

Model	Task	Source
🪐 StarCoder2	Code completion & repair	Hugging Face
⚙️ CodeT5+	Code understanding & generation	Hugging Face

Models and configurations are defined in:

src/config/model_config.yaml
src/config/training_args.yaml

---

🧩 Usage

🔹 Run the FastAPI Backend

uvicorn src.app.main:app --reload

Access it at: http://localhost:8000/docs

🔹 Launch Gradio Web Demo

python deployment/huggingface_spaces/app.py

Then open the local or Hugging Face demo URL.

---

🧪 Example API Request

curl -X POST "http://localhost:8000/fix" -H "Content-Type: application/json" -d '{"code": "def add(a,b)
 return a+b"}'

Response:

{
  "fixed_code": "def add(a, b):
    return a + b",
  "explanation": "Added missing colon and indentation."
}

---

📊 Training & Evaluation

1. Data Exploration: notebooks/01_data_exploration.ipynb
2. Preprocessing: notebooks/02_preprocessing.ipynb
3. Model Finetuning: notebooks/03_model_finetuning.ipynb
4. Evaluation: notebooks/04_evaluation.ipynb
5. Inference Demo: notebooks/05_inference_demo.ipynb

All experiment outputs are stored under:

experiments/logs/
experiments/checkpoints/
experiments/results/