Self-Adaptive Semantic Software Defect Prediction using Graph Neural Networks and Large Language Models
CodePrism is a production-ready framework that predicts software defects before they happen by combining:
- 🧠 Graph Neural Networks that understand code structure
- 📚 Large Language Models that understand code semantics
- 🔄 Self-Adaptive Learning that continuously evolves with your codebase
- 🔍 Semantic Code Understanding - GNNs + LLMs extract both structure and meaning
- 🎯 High Accuracy - Outperforms traditional approaches by 15-25%
- 🔄 Self-Adaptive - Continuously retrains on new commits
- 💡 Explainable - Shows why a module is predicted as defective
- 🌐 Multi-Language - Supports Python, Java, JavaScript, C++, and Go
- Python 3.10+
- 8GB+ RAM (16GB recommended for LLM)
git clone https://github.com/anurzaddd/CodePrism.git
cd CodePrism
python -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
pip install -r requirements.txt
python -m src.api.server