🧠 NeuroCode

NeuroCode is an AI-powered, full-stack code intelligence platform designed for proactive bug prediction, optimization, translation, documentation, and developer assistance. It uses two specialized fine-tuned LLMs:

• 🔍 WizardCoder-7B: Bug detection, code optimization, and translation
• 🧾 CodeLlama-Instruct-7B: Documentation generation and chatbot functionality

NeuroCode aims to empower developers with intelligent automation and project-specific insights, fully integrated into a modern web platform.

---

🚀 Features

✅ Proactive Bug Prediction using fine-tuned WizardCoder on BugSwarm & Defects4J ✅ Code Optimization trained on CodeXGLUE + HumanEval ✅ Code Translation using TransCoder dataset (Python ↔ Java ↔ C++) ✅ Documentation Generation using CodeSearchNet ✅ Developer Chatbot trained on GitHub Issues via API ✅ Real-time Collaborative Debugging (WebSocket enabled) ✅ Security Vulnerability Detection via custom scanner ✅ Analytics Dashboard (bug fix stats, usage logs, model trends) ✅ Fully integrated RAG module (FAISS + GitHub context) ✅ RLHF-ready feedback and simulated reward integration

---

🧩 File Structure

See the Final NeuroCode Structure for the complete project layout.

---

⚙️ Setup Instructions

🔧 Backend (Django + DRF + WebSockets)

cd backend
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
python manage.py migrate
python manage.py runserver

⚛️ Frontend (React)

cd frontend
npm install
npm start

🤖 Inference Server (FastAPI with Dual Models)

cd model/inference
uvicorn inference_router:app --reload --port 5000

Ensure checkpoints/wizardcoder_final and checkpoints/codellama_final directories exist with your trained models.

---

📦 Model Training

WizardCoder-7B

Trained on:

• BugSwarm, Defects4J (Bug Prediction)
• CodeXGLUE, HumanEval (Optimization)
• TransCoder (Code Translation)

cd model
python datasets/preprocess_wizardcoder.py
python finetuning/train_wizardcoder_final.py

CodeLlama-Instruct-7B

Trained on:

• CodeSearchNet (Documentation)
• GitHub Issues API (Chatbot)

python datasets/fetch_github_issues.py
python datasets/preprocess_codellama.py
python finetuning/train_codellama_final.py

---

🐳 Deployment

cd deployment
bash deploy.sh  # or use docker-compose up --build

Cloud Support:

• AWS EC2 with aws/ec2-setup.md
• Kubernetes with k8s/neurocode-deployment.yaml
• CI/CD with Jenkins (Jenkinsfile)

---

🧪 Testing

pytest tests/backend_tests
pytest tests/load_tests
npm run cypress:e2e

---

📚 Documentation

• docs/whitepaper.md: Research insights and architecture
• docs/paper.latex: Publication-ready format
• docs/presentation_outline.md: Slide draft for demo/pitch

---

🔒 License

MIT License – see LICENSE

---

💬 Credits

Built by a passionate team of 3 for our Advanced LLM Capstone. Dual-models, full-stack pipeline, and research-ready design.

Mentor: [Your Mentor Name] Contributors: [You], [Team Member 2], [Team Member 3]

---

❤️ Acknowledgments

• WizardCoder
• CodeLlama
• Hugging Face, PyTorch, Django, React, FastAPI, FAISS, GitHub API