📄 RAG Document Intelligence

Production-grade RAG + OCR pipeline for intelligent document processing. Upload PDFs or scanned images, ask questions in natural language, and extract structured data — all running locally with zero cost.

Built for processing German and English documents: invoices, contracts, university forms, bureaucratic paperwork.

---

What it does

Feature	Description
OCR	Extracts text from native PDFs and scanned documents (Tesseract, English + German)
RAG Q&A	Ask questions about your documents — answers grounded in actual content with source citations
Structured Extraction	Automatically finds dates, amounts (EUR/USD), emails, phone numbers, IBANs, invoice numbers
Multi-document	Upload multiple documents and query across all of them
Bilingual	Handles both German and English documents natively

---

---

Tech Stack

Layer	Technology	Why
LLM	Groq (Llama 3.1 8B)	Free API, fast inference, no GPU needed
Embeddings	sentence-transformers/all-MiniLM-L6-v2	Runs locally, no API key, 384-dim vectors
Vector Store	FAISS	Facebook's similarity search, in-memory, fast
Framework	LangChain	Industry-standard RAG orchestration
OCR	Tesseract + pdfplumber	Open-source, supports 100+ languages
Backend	FastAPI	Async, auto-docs at /docs, production-ready
Frontend	Streamlit	Rapid prototyping, built-in chat components
Infra	Docker + docker-compose	One-command deployment

---

Quick Start

Option 1: Docker (recommended)

# Clone
git clone https://github.com/GouravJr/rag-document-intelligence.git
cd rag-document-intelligence

# Configure
cp .env.example .env
# Edit .env → add your GROQ_API_KEY (free at https://console.groq.com)

# Run
docker-compose up --build

# Open
# API docs: http://localhost:8000/docs
# Frontend: http://localhost:8501

Option 2: Local Development

# Prerequisites: Python 3.11+, Tesseract OCR
# macOS: brew install tesseract tesseract-lang poppler
# Ubuntu: sudo apt install tesseract-ocr tesseract-ocr-deu poppler-utils
# Windows: download from https://github.com/UB-Mannheim/tesseract/wiki

# Clone & setup
git clone https://github.com/GouravJr/rag-document-intelligence.git
cd rag-document-intelligence
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt

# Configure
cp .env.example .env
# Edit .env → add your GROQ_API_KEY

# Terminal 1: Start API
uvicorn app.main:app --reload

# Terminal 2: Start Frontend
streamlit run streamlit_app/app.py

# Open http://localhost:8501

Get a free Groq API key

1. Go to console.groq.com
2. Sign up (free, no credit card)
3. Create an API key
4. Paste it in your .env file

---

API Reference

Once running, full interactive docs at http://localhost:8000/docs

Endpoint	Method	Description
/upload	POST	Upload PDF/image → OCR → ingest into RAG
/query	POST	Ask a question about uploaded documents
/documents	GET	List all ingested documents
/extract/{doc_id}	GET	Get structured data extraction
/clear	DELETE	Reset everything
/health	GET	Health check

---

Project Structure

rag-document-intelligence/
├── app/
│   ├── config.py          # Configuration & env vars
│   ├── main.py            # FastAPI endpoints
│   ├── ocr.py             # PDF & image text extraction
│   ├── rag.py             # RAG pipeline (chunk → embed → retrieve → generate)
│   └── extraction.py      # Structured data extraction (regex + heuristics)
├── streamlit_app/
│   └── app.py             # Streamlit chat + upload interface
├── tests/
│   └── test_pipeline.py   # Unit tests for extraction & pipeline
├── Dockerfile
├── docker-compose.yml
├── requirements.txt
├── .env.example
└── README.md

---

How it works

1. Document Ingestion

• Native PDFs: text extracted directly with pdfplumber (fast, accurate)
• Scanned PDFs/images: pages converted to images → Tesseract OCR (300 DPI, preprocessed)
• Hybrid mode: automatically detects which pages need OCR

2. RAG Pipeline

• Text split into overlapping chunks (1000 chars, 200 overlap)
• Chunks embedded with sentence-transformers/all-MiniLM-L6-v2 (runs on CPU)
• Stored in FAISS index for sub-millisecond similarity search
• On query: top-4 relevant chunks retrieved → sent to Groq LLM with custom prompt → grounded answer returned

3. Structured Extraction

• Regex-based extraction of dates, amounts, emails, phones, IBANs, reference numbers
• Supports German date formats (DD.MM.YYYY) and currency notation (1.234,56 EUR)
• Key-value pair detection for form-like documents
• Automatic language detection (German vs English)

---

Running Tests

pytest tests/ -v

---

Future Improvements

• Add table extraction from PDFs (with camelot or tabula)
• Fine-tune embedding model on German documents
• Add authentication for multi-user deployment
• Implement conversation memory for follow-up questions
• Add Kubernetes deployment manifests
• Support for XLSX and DOCX input

---

Author

Gourav Srinivasalu M.Sc. Artificial Intelligence Engineering · Universität Passau

• GitHub: github.com/GouravJr
• LinkedIn: linkedin.com/in/gourav-srinivasalu
• Email: gourav.srinivasalu@gmail.com

---

License

MIT License — see LICENSE for details.