README.md

metadata

title: ALDDS
emoji: 💻
colorFrom: purple
colorTo: red
sdk: docker
pinned: false

⚖️ Automated Legal Document Digitization System (ALDDS)

Upload a photo of a legal document (bailable warrant, summon, etc.) and seamlessly convert it into structured JSON data. ALDDS handles the OCR, AI parsing, secure storage, and live dispatcher notifications in a single streamlined pipeline—100% locally and offline.

🚀 Pipeline & Architecture

Image Upload → Cloudinary Hosting → PaddleOCR → Local Qwen2.5-1.5B LLM → MongoDB → WhatsApp Alert

1. OCR Extraction: Uses PaddleOCR to extract raw text from image uploads with high accuracy on complex legal layouts.
2. AI Parsing: Leverages a quantized Local Qwen2.5-1.5B GGUF model via llama-cpp-python to parse the unstructured OCR text into a strict 10-field JSON schema. Zero API costs, 100% data privacy.
3. Secure Storage: Automatically commits the digital record to a MongoDB database.
4. Live Dashboard: A real-time, searchable Police Dashboard built directly into the UI.
5. Instant Notifications: Dispatchers can notify Investigating Officers (IOs) instantly via zero-cost WhatsApp wa.me links populated from an internal CSV database.

---

📋 Features

• 100% Offline Local AI: Uses llama-cpp-python to run a highly optimized Qwen2.5-1.5B model directly on CPU. No external API keys (like OpenAI or NVIDIA) are needed, guaranteeing zero data leaks and no per-request costs.
• HF Free Tier Optimized: Specifically configured (non-streaming, 1.5B model, capped threads) to run quickly on Hugging Face's free 2 vCPU tier without timing out.
• Auto-Healing OCR: Includes an automated fix that detects and downgrades broken PaddlePaddle 3.x versions to the stable v2.6.2 runtime during startup, preventing CPU crashes.
• 👮 Live Police Dashboard: A dedicated tab for station dispatchers to monitor incoming warrants. Includes a real-time MongoDB search filter (by Case No, IO Name, Station, etc.).
• 💬 WhatsApp Integration: Zero-API-cost notifications. Selecting an officer from the dropdown dynamically pulls their phone number from officers.csv and opens a pre-filled WhatsApp window for the dispatcher to send manually.
• Strict Hallucination Guard: Custom Python post-processing runs regex checks to ensure the LLM didn't fabricate names, dates, or case numbers that aren't explicitly in the OCR text.

---

🔑 Environment Setup

To run ALDDS locally or in the cloud, you only need the following API keys configured in a .env file (or as Secrets on Hugging Face). No LLM API key is required.

# Cloudinary (Image Hosting)
CLOUDINARY_CLOUD_NAME=your_cloud_name
CLOUDINARY_API_KEY=your_api_key
CLOUDINARY_API_SECRET=your_api_secret

# MongoDB (Database)
MONGODB_URI=mongodb+srv://username:password@cluster0.../?retryWrites=true&w=majority

Variable	Where to get it
CLOUDINARY_*	Cloudinary Console → Dashboard
MONGODB_URI	MongoDB Atlas → Database → Connect

---

💻 Local Installation Guide

ALDDS handles all OCR and AI natively in Python. You do not need to install Tesseract or any system-level C++ compilers.

# 1. Clone repository
git clone https://huggingface.co/spaces/LovnishVerma/ALDDS
cd ALDDS

# 2. Install Dependencies
pip install -r requirements.txt

# 3. Run the Server
python app.py

The interface will launch at http://127.0.0.1:7860.

Note: The first launch automatically downloads the Qwen2.5-1.5B GGUF model (~1.1 GB) and the PaddleOCR weights. Subsequent launches will use the cached files.

---

☁️ Hugging Face Spaces Deployment

If you are deploying ALDDS to Hugging Face Spaces, follow these steps:

1. Set Up Secrets: Go to your Space Settings → Variables and secrets. Add CLOUDINARY_CLOUD_NAME, CLOUDINARY_API_KEY, CLOUDINARY_API_SECRET, and MONGODB_URI.
2. Ensure officers.csv: Upload your officers.csv file (with Officer_Name and Phone_Number columns) to the root of the repository.
3. Restart: Always click Factory Rebuild or Restart Space after adding Secrets so the new environment variables are loaded into the container.

---

📊 Extracted JSON Schema

The LLM is strictly prompted to extract the following fields. If a field cannot be explicitly found in the text, it defaults to null.

JSON Key	Description
Case_FIR_Number	FIR or court case reference number
Act_and_Sections	Applicable IPC/CRPC legal acts and sections
Type_of_Document	Warrant, summon, notice, etc.
Target_Police_Station	Police station the document is addressed to
IO_Name_and_Belt_No	Investigating Officer's name and belt number
IO_Mobile_Number	IO's contact number
Person_Name_To_Serve	Name of the person to be served/arrested
Person_Address	Address of the target person
Court_Name	Issuing court name
Hearing_Date	Scheduled hearing or appearance date (DD-MM-YYYY)

---

Built to streamline station workflows, eliminate manual data entry, and instantly notify field officers—all while keeping sensitive legal data entirely offline and private.