🧠 Smart Contract Audit Agent – LLM-Powered Bug Detection

An AI-powered vulnerability detection tool for Solidity smart contracts. Built using Hugging Face Transformers and trained on the SmartBugs dataset, this system uses Microsoft’s Phi‑3‑Mini‑4K‑Instruct model to analyze and identify bugs and flaws in Solidity code. The interface is built with Streamlit, enabling developers to paste snippets and receive instant feedback with explanations and suggestions. The core experiments were developed and run in Google Colab.

---

✨ Key Features

• 🔍 SmartBugs Dataset
  Trained using a curated dataset of Solidity smart contract snippets annotated with vulnerabilities categorized under the DASP taxonomy (e.g., reentrancy, access control, bad randomness) :contentReference[oaicite:1]{index=1}.

• 🤖 Phi‑3‑Mini‑4K‑Instruct Model
  A 3.8-billion-parameter instruction-tuned model optimized for reasoning, code analysis, and chat formats—available via Hugging Face and Azure :contentReference[oaicite:2]{index=2}.

• 💻 Streamlit UI for Code Auditing
  A lightweight, interactive interface that lets users paste Solidity code and receive structured feedback about vulnerabilities and mitigation suggestions.

• 🧪 Colab-Based Development
  Experiments, training, and inference all executed on Google Colab, making the project reproducible and easy to run.

---

🧩 Project Structure

smart-contract-audit-agent/
│
├── notebooks/               # Google Colab notebooks (training, inference, demos)
├── README.md
└── LICENSE

---

🛠️ Getting Started

1. Clone the Repository

git clone https://github.com/ai-art-dev99/smart-contract-audit-agent.git

2. Set Up Virtual Environment & Install Dependencies

run on google colab
A100 GPU

3. Run the Streamlit UI

streamlit run streamlit_app.py

This will open a local web interface where Solidity code can be pasted for analysis.

4. Use the Colab Notebook

Open the included Google Colab notebook in notebooks/ to train the Phi‑3‑Mini‑4K‑Instruct model on the SmartBugs dataset, run inference, and generate sample reports.

---

🧠 Technologies Used

Component	Description
Dataset	SmartBugs curated contracts with tagged Solidity vulnerabilities (e.g. reentrancy, DASP taxonomy) (GitHub, smartbugs.github.io, GitHub)
Language Model	Microsoft Phi‑3‑Mini‑4K‑Instruct (3.8B params, tuned for code reasoning) (Hugging Face, Microsoft Azure)
Interface	Streamlit for interactive UI
Platform	Google Colab for development, training & inference
Libraries	transformers, torch, streamlit, other required dependencies

---

💡 Future Improvements

• Add fine-tuning options for other Phi‑3 variants (e.g., 128K context window version)
• Extend dataset with real-world contracts or vulnerabilities beyond SmartBugs
• Add file upload / history / exportable report features in UI
• Incorporate ML-based explanation generation and severity scoring

---

📄 License

Licensed under the MIT License. See the LICENSE file for details.

---

🙏 Acknowledgements

• SmartBugs project—providing a standard benchmark dataset for smart contract vulnerability detection (Microsoft Azure, GitHub, Windows Central)
• Microsoft Phi‑3 family of instruction‑tuned models, optimized for compact hardware and high reasoning accuracy (Hugging Face)
• Hugging Face and Streamlit for their versatile tooling

---

Built to empower smart contract developers with AI‑driven security insights.