| <<<<<<< HEAD | |
| # Risk-clause-classifiers | |
| ======= | |
| --- | |
| license: mit | |
| --- | |
| >>>>>>> 17efa33e6a905f74adced154117a7489f1672bab | |
| # Risk Clause Classifier | |
| A lightweight AI system for classifying legal contract clauses into: | |
| - Clause Category (10 classes) | |
| - Risk Level (Low / Medium / High) | |
| This project is designed for low-resource environments and CPU-based execution. | |
| ## π Project Overview | |
| The system uses a distilled domain-specific language model to generate clause embeddings and applies lightweight classifiers for: | |
| - Category Prediction | |
| - Risk Assessment | |
| The goal is to build an efficient, scalable, and hackathon-ready legal clause analysis tool. | |
| ## π Architecture | |
| Clause Text | |
| β Transformer Embeddings | |
| β Category Classifier | |
| β Risk Level Classifier | |
| ## β Tech Stack | |
| - Python | |
| - Transformers | |
| - Hugging Face Hub | |
| - Scikit-learn | |
| - FastAPI | |
| - Git & GitHub | |
| ## π Repository Structure | |
| - `/src` β Training & inference scripts | |
| - `/models` β Saved classifier models | |
| - `/app.py` β API server | |
| - `requirements.txt` β Dependencies | |
| ## π Deployment | |
| The model files are hosted on Hugging Face. | |
| Codebase is maintained on GitHub for team collaboration. | |
| ## π₯ Team Collaboration | |
| - GitHub β Code management | |
| - Hugging Face β Model storage | |
| - Multi-device sync enabled via Git remotes | |
| ## π Hackathon Focus | |
| - CPU-friendly design | |
| - Lightweight inference | |
| - Structured JSON output | |
| - Legal domain specialization | |
| --- | |
| Built for efficient legal risk clause analysis. | |