A Premium, Agentic Document Intelligence Platform that combines Traditional OCR with Generative AI Vision.
This project was designed and developed by students from Nutan College of Engineering and Research, Talegaon Dhabade, Pune under Intel Unnati Industrail Training 2025 Programme. Feel free to connect with us and explore our work!
A huge thank you to our mentor for their guidance, insights, and continuous support throughout the development of this project.
Prof. Priyanka Vyas
| Landing Page | Dashboard & Chat |
|---|---|
 |
 |
| Inspector | Knowledge Base |
|---|---|
 |
 |
👉 https://youtu.be/nZSyUKTinMs
This video demonstrates:
- End-to-end document ingestion
- OCR vs Gemini Vision switching
- Visual RAG responses
- Inspector & Knowledge Base auditing
The Intel Enterprise Document Analyzer is designed to solve the "Last Mile" problem of document intelligence: extracting structured data from unstructured, messy real-world PDFs. By dynamically switching between Tesseract OCR (for speed) and Google Gemini Vision (for reasoning), it achieves high accuracy even on handwritten or complex documents.
📘 Read the Full User Guide
📘 Read the System Architecture
📘 Read the API Reference
- Hybrid Extraction Engine: Automatically handles Digital vs. Scanned vs. Handwriting PDFs
- Deep Artifact Extraction: Isolates tables into DataFrames and crops images for separate indexing
- Visual RAG: Search results provide not just text, but visual "evidence" crops from the original PDF
- Agentic Inspector: Dedicated UI to audit every chunk, image, and table found in a document
- Enterprise-Ready Pipeline: Modular ingestion, parsing, chunking, embedding, and retrieval
- Python 3.9+
- Docling OCR
- Google Gemini API Key
-
Clone & Install
git clone <repo> cd IntelProject pip install -r requirements.txt
-
Configure API Key
Copy code export GOOGLE_API_KEY="your_api_key" # or configure inside backend/config.py Run the System (Two Terminals)
-
Backend (FastAPI):
Copy code uvicorn backend.main:app --host 127.0.0.1 --port 8000 Frontend (Streamlit):
Copy code streamlit run frontend/app.py
The backend expects the following structure:
data/
├── processed/
├── uploads/
├── vectordb/
└── static/
├── images/
├── pages/
└── pdfs/
Create manually if needed:
mkdir -p data/processed \
data/static/images \
data/static/pages \
data/static/pdfs \
data/uploads \
data/vectordb
FastAPI Backend: High-performance async API
Streamlit Frontend: Interactive enterprise dashboard
Vector Store: Persistent document embeddings
Vision + Text RAG: Multi-modal retrieval with visual grounding
Inspector Mode: Transparent AI auditing
API keys are never hardcoded (recommended via environment variables)
Containers run isolated via Docker network
Designed for private VM / enterprise network usage
User Guide : Detailed walkthrough of features
System Architecture : Technical deep-dive
API Reference : Backend endpoints
MIT License
Give it a star ⭐ and feel free to fork or extend it for research, enterprise pilots, or hackathons.