Document Intelligence System

Overview

The Document Intelligence System is an intelligent, scalable pipeline designed to ingest scanned documents or photographs of physical documents, automatically classify them, and extract structured information. It features a modern, user-friendly interface that allows users to review the AI's confidence scores and correct any extracted data before final confirmation.

Key Features

• Document Upload: Secure, drag-and-drop interface capable of handling multiple complex documents at once (PDF, JPEG, PNG, WEBP).
• Automatic Document Classification: Dynamically identifies the document type (e.g., Driver's License, Passport, Vehicle Power of Attorney).
• OCR-based Text Extraction: Utilizes state-of-the-art Computer Vision engines to extract raw text, even from blurry or complex images.
• Structured Data Extraction: Employs Large Language Models (LLMs) to map the raw OCR text perfectly into strict JSON schemas.
• User Review and Correction Interface: Clean dual-pane validation UI for human-in-the-loop review.
• Metadata Storage: Tracks document processing status, AI reasoning confidence, and completion percentages.
• Secure Document Storage: Immutable S3 Object Storage integration for raw document assets.

Technology Stack

• Frontend: React.js, Vite, Framer Motion (Animations), Axios, Vanilla CSS
• Backend: Python 3, FastAPI, Uvicorn, Motor (Async MongoDB), Tenacity (Backoff Retries)
• AI / OCR: docTR (Computer Vision OCR), Google Gemini 3 Flash Preview (LLM Data Restructuring), PyMuPDF
• Storage: MongoDB (Document Metadata), IDrive e2 / S3-Compatible Storage (Blob Storage)

Repository Structure

.
├── backend/          # FastAPI server, AI pipeline scripts, S3 connections, and MongoDB Models
├── frontend/         # React application, Vite config, UI components, and styles
├── docs/             # Technical specifications, implementation plans, and architecture designs
└── docker-compose.yml# Production container orchestration

Setup Instructions

Option 1: Docker (Recommended)

The easiest way to run the application is using the provided Docker Compose configuration, which automatically sets up the Frontend, Backend, and MongoDB databases in isolated containers.

1. Set up Environment Variables: Ensure your backend/.env file is properly configured with your API keys (see Environment Variables section below).

2. Deploy the fleet:

   docker compose up --build -d

Option 2: Run Locally (Manual Setup)

1. Start MongoDB: Ensure you have a local instance of MongoDB running on port 27017.

2. Run the Backend:

cd backend
python3 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate
pip install -r requirements.txt
uvicorn main:app --reload --host 0.0.0.0 --port 8000

3. Run the Frontend:

cd ../frontend
npm install
npm run dev

Environment Variables

To securely run this application, create a .env file in the backend/ directory with the following configuration keys:

# S3 / IDrive e2 Storage Configuration
AWS_ACCESS_KEY_ID="your_access_key"
AWS_SECRET_ACCESS_KEY="your_secret_key"
AWS_REGION="us-west-2"
S3_BUCKET_NAME="your-bucket-name"
S3_ENDPOINT_URL="https://s3.us-west-2.idrivee2.com"

# Google Gemini API
GEMINI_API_KEY="your_gemini_api_key_here"

# Database (Optional: defaults to localhost if running manually)
MONGODB_URL="mongodb://localhost:27017"

Running the Application

Once the services are booted (either via Docker or manually), the application will be accessible directly from your browser:

• Frontend UI: Navigate to http://localhost:5173
• Backend API Sandbox: Navigate to http://localhost:8000/docs to test endpoints natively via Swagger UI.

Example Workflow

1. Upload Document: Click or drag a scanned file into the dropzone. It is securely saved to S3.
2. Extraction: The FastAPI background worker routes the image through the docTR computer vision engine, grabs the messy text, and passes it to the Gemini 3 LLM layer to cleanly extract JSON according to a strict template.
3. Review: The User Interface dynamically renders the returned JSON key-values next to the original raw document.
4. Confirm: The user can safely edit the AI's predictions and successfully Accept the data, pushing it cleanly into MongoDB.

Additional Documentation

• Architecture and Scale: Detailed system design, schema enforcement logic, architecture decisions, and scalability considerations for Enterprise loads are strictly documented inside docs/DESIGN.md.
• System Integrations: A conceptual breakdown mapping the pipeline within a strict Identity Verification environment is documented inside docs/CONCEPTUAL.md.