⚖️ Law-Lens — AI-Powered Legal Document Analysis

Law-Lens is an intelligent legal document analysis backend that extracts, interprets, and visualizes key insights from legal PDFs. It combines legal-domain NLP models, LLMs, vector search, and OCR to help legal professionals quickly understand and navigate complex documents.

---

Key Features

Feature	Description
📄 Smart PDF Extraction	Extracts text from any PDF — born-digital (fast path) or scanned (OCR fallback)
🔑 Legal Keyword Extraction	Domain-aware keyword extraction using LegalBERT + KeyBERT
📖 Keyword Definitions	Plain-English, one-sentence meanings for every legal term via LLM
⚖️ Indian Case Law Mapping	Maps keywords to IPC/Act sections with IndianKanoon search links
🎯 Paragraph Importance Scoring	Hybrid semantic filter + LLM scoring — tags paragraphs as High / Medium / Low
🖍️ Highlighted PDF Export	Color-annotated PDF (red = high, yellow = medium) with bounding-box precision
💬 RAG-Powered Chatbot	Ask questions about any uploaded document — grounded in its content via FAISS retrieval
🚀 FastAPI Backend	Clean REST API with interactive Swagger docs at /docs

---

Tech Stack

Python 3.8+  |  FastAPI  |  Uvicorn
────────────────────────────────────
AI / NLP:    LegalBERT · KeyBERT · SentenceTransformers · Groq Llama 3.1
Vector:      FAISS (IndexFlatL2)
OCR:         Tesseract · pdf2image · Poppler
PDF:         pypdfium2 · PyMuPDF · PyPDF2 · Pillow
API:         FastAPI · Pydantic · python-multipart
UI (Alt):    Streamlit

---

Architecture

┌─────────────┐     ┌──────────────────┐     ┌──────────────────────┐
│  FastAPI     │────▶│  Service Layer   │────▶│   Domain Modules     │
│  Routes      │     │  (Orchestration) │     │  (Pure Logic)        │
│              │     │                  │     │                      │
│  health.py   │     │  pdf_service.py  │     │  pdf_processor.py    │
│  pdf.py      │     │  chat_service.py │     │  keyword.py          │
│  chat.py     │     │  json_utils.py   │     │  keyword_meaning.py  │
└─────────────┘     └──────────────────┘     │  semantic_importance │
                                              │  vector_store.py     │
                                              │  chatbot.py          │
                                              │  highlight_pdf.py    │
                                              │  case_law_fetcher.py │
                                              │  ocr.py              │
                                              └──────────────────────┘

---

Processing Pipeline

PDF Upload
    │
    ▼
┌──────────────────────────────────────────────┐
│ 1. Text Extraction                           │
│    Fast: pypdfium2  │  Fallback: PyPDF2      │
│    OCR:  Tesseract (if < 50 chars extracted) │
└──────────────────────────────────────────────┘
    │
    ▼
┌──────────────────────────────────────────────┐
│ 2. Legal Keyword Extraction (LegalBERT)      │
└──────────────────────────────────────────────┘
    │
    ▼
┌──────────────────────────────────────────────┐
│ 3. Keyword Meanings  (Groq Llama, JSON mode) │
└──────────────────────────────────────────────┘
    │
    ▼
┌──────────────────────────────────────────────┐
│ 4. Case Law Mapping  (Groq → IndianKanoon)   │
└──────────────────────────────────────────────┘
    │
    ▼
┌──────────────────────────────────────────────┐
│ 5. Paragraph Importance Scoring              │
│    Hybrid: SentenceTransformer filter        │
│           + Groq Llama batch scoring         │
│           + Keyword-count fallback           │
└──────────────────────────────────────────────┘
    │
    ▼
┌──────────────────────────────────────────────┐
│ 6. FAISS Indexing (sliding-window chunks)    │
└──────────────────────────────────────────────┘
    │
    ▼
┌──────────────────────────────────────────────┐
│ 7. PDF Highlighting (PyMuPDF color annot.)   │
└──────────────────────────────────────────────┘
    │
    ▼
 Response with session_id + analysis data

---

Quick Start

Prerequisites

• Python 3.8+
• Tesseract OCR
• Poppler
• A Groq API key

Setup

# Clone and navigate
cd Law-Lens

# Install dependencies
pip install -r requirements.txt

# Create .env file
cat > .env << EOF
GROQ_API_KEY=your_groq_api_key
TESSERACT_PATH=C:/path/to/tesseract.exe
POPPLER_BIN_PATH=C:/path/to/poppler/bin
EOF

# Start the API server
python run_api.py

The server starts at http://localhost:8000. Interactive docs at http://localhost:8000/docs.

Streamlit UI (Alternative)

streamlit run app.py

---

API Endpoints

Method	Endpoint	Description
GET	/	Root health check
GET	/health	Server status + model load state
POST	/pdf/upload	Upload & analyze a PDF document
GET	/pdf/session/{session_id}	Retrieve analysis data for a session
DELETE	/pdf/session/{session_id}	Delete session & cleanup files
GET	/pdf/download/highlighted/{session_id}	Download highlighted PDF
GET	/pdf/download/keywords/{session_id}	Download keywords as .txt
GET	/pdf/download/text/{session_id}	Download raw extracted text
POST	/chat/	Ask a question about a document

---

Project Structure

Law-Lens/
├── main.py                        # FastAPI entry point
├── app.py                         # Streamlit UI entry point
├── run_api.py                     # Server launcher
├── requirements.txt               # Python dependencies
├── config.toml                    # Streamlit file-watcher config
├── .env                           # API keys & system paths
│
├── routes/                        # HTTP layer
│   ├── health.py                  # Health check endpoints
│   ├── pdf.py                     # PDF upload / download / session
│   └── chat.py                    # Chat endpoint
│
├── services/                      # Business logic & orchestration
│   ├── pdf_service.py             # PDF pipeline orchestrator + state
│   ├── chat_service.py            # Chat orchestration
│   └── json_utils.py              # NumPy-safe JSON serialization
│
├── modules/                       # Domain logic (no HTTP awareness)
│   ├── pdf_processor.py           # PDF text extraction & cleaning
│   ├── keyword.py                 # LegalBERT keyword extraction
│   ├── keyword_meaning.py         # Groq-based keyword definitions
│   ├── semantic_importance.py     # Hybrid paragraph scoring
│   ├── vector_store.py            # FAISS index creation & search
│   ├── chatbot.py                 # RAG Q&A pipeline
│   ├── highlight_pdf.py           # PDF color annotation
│   ├── case_law_fetcher.py        # Indian legal section mapper
│   ├── ocr.py                     # Tesseract OCR pipeline
│   └── utils/
│       └── text_cleaner.py        # Keyword normalization
│
└── test_new_api.py                # Integration test suite

---

Design Highlights

• Resilient fallbacks — PDF extraction degrades gracefully from fast → slow → OCR; paragraph scoring falls back to keyword heuristics if LLM unavailable; highlighting skips gracefully on failure
• API cost optimization — semantic filter reduces LLM calls by ~60-80% by pre-filtering non-legal paragraphs before Groq scoring
• Robust PDF matching — token overlap + fuzzy ratio heuristics handle layout variations in legal PDFs
• Lazy model loading — heavy ML libraries (PyTorch, Transformers) load on-demand, not at import time
• NumPy-safe serialization — recursive converter ensures FAISS/Numpy types don't break JSON responses

---

Testing

# Start the server first, then:
python test_new_api.py

The integration test validates: health check → PDF upload → chat → downloads → session management.

---

Future Roadmap

• Persistent storage (PostgreSQL + Redis) for production scalability
• Asynchronous LLM calls for parallel pipeline stages
• WebSocket-based progress streaming
• Docker containerization with multi-stage build
• CI/CD pipeline (GitHub Actions → cloud deploy)
• JWT authentication & rate limiting
• Unit test coverage with mocked LLM responses