⚖️ IPC Predictor – Multilingual Legal AI Assistant

📌 Overview

This project is an AI-powered legal assistant that predicts the most relevant Indian Penal Code (IPC) sections from a user-provided crime description.
It supports 10+ Indian languages, automatically translates queries to English, and provides detailed legal information including offense, punishment, cognizability, and bailability.

---

✨ Key Features

• 🔍 Hybrid IPC Prediction → Combines Sentence-BERT embeddings + TF-IDF keyword matching
• 📊 Severity Calibration → Aligns crime severity vs punishment severity for more realistic predictions
• 🌍 Multilingual Support → Handles queries in English, Hindi, Telugu, Tamil, Malayalam, Kannada, Marathi, Bengali, Gujarati, Punjabi, Odia, Urdu using fastText + Meta NLLB-200
• ⚡ Real-Time Predictions → Precomputed embeddings and vector caching for fast results
• 🖥️ Command-line Interface → Simple CLI to test and explore predictions

---

🛠️ Tech Stack

• Programming: Python
• Libraries: Sentence-BERT, scikit-learn, spaCy, TensorFlow, Hugging Face Transformers, fastText
• Models:
  • SentenceTransformer: all-mpnet-base-v2
  • Translation: facebook/nllb-200-distilled-600M
  • Language Detection: fastText lid.176.bin

---

📂 Project Structure

IPC_predictor/ │── ipc.py # Core IPC prediction engine (hybrid + severity calibration) │── translator.py # Multilingual support (language detection + translation) │── new.py # CLI interface with confidence filtering │── test_import.py # Basic import test │── test_fixed_translator.py # Debugging translation issues │── debug_translation.py # Translation test script │── merged.json # IPC dataset (sections, punishments, details) │── ipc_embeddings.pkl # Precomputed embeddings │── ipc_tfidf.pkl # Precomputed TF-IDF vectors │── README.md # Project documentation

🚀 Usage

1️⃣ Clone the repository

git clone https://github.com/Visshu78/IPC_predictor.git
cd IPC_predictor

pip install -r requirements.txt
python -m spacy download en_core_web_sm

python new.py

python translator.py

📈 Future Enhancements

✅ Web/Streamlit interface for user-friendly demo ✅ Evaluation on labeled dataset for accuracy benchmarks ✅ Expand IPC dataset with more detailed case descriptions

Myself

Vishal Dhawal

🎓 B.Tech CSE @ IIIT Kottayam | Minor @ IIT Mandi 🔬 Interests: Machine Learning, Computer Vision, NLP, Human-Computer Interaction 🌐 GitHub: @Visshu78