LexTransition-AI

Live Demo: https://kvbgkvw4mehwhhdjt7crrg.streamlit.app/

LexTransition AI is an open-source, offline-first legal assistant. It helps users navigate the transition from old Indian laws (IPC/CrPC/IEA) to the new BNS/BNSS/BSA frameworks. Using local Machine Learning and OCR, it analyzes legal documents and maps law sections with 100% grounded accuracy.

---

⚖️ LexTransition AI: Law Mapper & Document Analyzer

LexTransition AI is an open-source, offline-first legal assistant. It helps users navigate the transition from old Indian laws (IPC/CrPC/IEA) to the new BNS/BNSS/BSA frameworks. Using local Machine Learning and OCR, it analyzes legal documents and maps law sections with 100% grounded accuracy.

---

🚀 Key Modules

• 🔄 The Law Transition Mapper: The core engine that maps old IPC sections to new BNS equivalents. It highlights specific changes in wording, penalties, and scope.
• 🖼️ Multimodal Document Analysis (OCR): Upload photos of legal notices or FIRs. The system extracts text using local OCR and explains "action items" in simple language.
• 📚 Grounded Fact-Checking: Every response is backed by official citations. The AI identifies the exact Section, Chapter, and Page from the official Law PDFs to prevent hallucinations.

---

🛠️ Offline Tech Stack (No-API Approach)

To ensure privacy and offline accessibility, this project can be configured to run without external APIs:

• Backend: Python, LangChain/LlamaIndex.
• OCR: EasyOCR or PyTesseract (Local engines).
• Vector DB: ChromaDB or FAISS (Local storage instead of Pinecone/Milvus).
• Local LLM: Llama 3 or Mistral via Ollama or LM Studio (Runs on your GPU/CPU).
• Frontend: Streamlit Dashboard.

---

📂 Project Structure

LexTransition-AI/
├── app.py                 # Streamlit UI
├── requirements.txt       # Local ML libraries
├── engine/
│   ├── ocr_processor.py   # Local OCR logic
│   ├── mapping_logic.py   # IPC to BNS mapping dictionary
│   └── rag_engine.py      # Local Vector Search logic
└── models/                # Local LLM weights (Quantized)

---

⚙️ Installation & Local Setup

Option A: Using Docker (Recommended)

The easiest way to run LexTransition-AI is with Docker. This handles all dependencies (including Tesseract OCR and system libraries) automatically.

1. Clone the repository:

   git clone [https://github.com/centiceron/LexTransition-AI.git](https://github.com/centiceron/LexTransition-AI.git)
   cd LexTransition-AI

2. Build the Docker Image

   docker build -t lextransition.

3. Run the Application

   docker run -p 8501:8501 lextransition

4. Open the App

   http://localhost:8501

---

Current Implementation Status

• Streamlit UI (app.py) — implemented (interactive pages for Mapper, OCR, Fact-check).
• OCR — local helper supporting EasyOCR and pytesseract (install system tesseract for pytesseract).
• IPC→BNS Mapping — in-memory mapping with fuzzy match; UI supports adding mappings at runtime.
• Grounded Fact-Check — simple PDF ingestion and page-level keyword search using pdfplumber (add PDFs to ./law_pdfs via UI).
• RAG/LLM & full offline guarantees — NOT implemented yet (placeholders/stubs present).

---

Quick Start (local)

• Install Python dependencies: pip install -r requirements.txt

• (Optional) Install Tesseract binary for pytesseract:

  • Ubuntu: sudo apt install tesseract-ocr
  • Mac (brew): brew install tesseract

• Launch: streamlit run app.py

To use Grounded Fact-Check, upload law PDFs in the Fact-Check page (or drop them into ./law_pdfs) and click "Verify with Law PDFs".

---

Persistence & Testing

• Mappings are persisted to mapping_db.json (in project root). You can add mappings in the UI; they are saved to this file.

• Run tests:

  • pip install -r requirements.txt
  • pytest -q

---

OCR Benchmark Harness

Use scripts/ocr_benchmark.py with a CSV dataset (image_path,ground_truth) to compute:

• Character Error Rate (CER)
• Keyword Recall

Example:

• python scripts/ocr_benchmark.py --dataset data/ocr_dataset.csv --report ocr_report.md

Optional features (embeddings & local LLM)

Embedding-based RAG (FAISS + sentence-transformers)

• Install (optional): pip install sentence-transformers numpy faiss-cpu
• Enable: export LTA_USE_EMBEDDINGS=1
• Index persists in ./vector_store

---

Local LLM integration (Ollama)

• Configure: export LTA_OLLAMA_URL=http://localhost:11434
• The app will use this endpoint for better plain-language summaries.

---

CI

• A GitHub Actions workflow (lextransition-ci.yml) runs pytest for the project on PRs.

---

Next Steps / TODO

• Replace page-level keyword search with embeddings + vector store (Chroma/FAISS) + exact citation offsets.
• Add persistent mapping DB + import tools for official IPC→BNS mappings.
• Integrate local LLM for summaries/explanations (Ollama / LM Studio).
• Add tests and CI for engine modules.