📺 Watch the full end-to-end demo featuring Asynchronous PDF ingestion, LangGraph agentic routing, and 100% test coverage validation.

DocuMind is a production-grade, containerized RAG (Retrieval-Augmented Generation) Knowledge Management System. It mimics a secure Azure Enterprise setup, featuring an agentic core that intelligently routes user queries between general conversation and strict document search.

DocuMind-Enterprise is a production-grade Reference Architecture for building secure, compliant Retrieval-Augmented Generation (RAG) systems.

Most RAG demos fail in enterprise production because they lack governance and cost control. This project implements a strict "Citation-First" architecture designed for regulated industries (Legal, Finance, GDPR-compliant sectors). It enforces:

1. Strict Source Attribution: No answer is generated without a verified PDF page reference (Zero Hallucination Policy).
2. Agentic Routing: Uses LangGraph to intelligently distinguish between "general chitchat" and "database queries," significantly reducing token costs and latency.
3. Asynchronous Ingestion: Non-blocking FastAPI pipelines for high-throughput document processing.

The application is built on a Microservices architecture using Docker Compose:

1. Frontend (React + Vite): A modern "Glassmorphism" UI with streaming chat support and file ingestion status tracking.
2. Backend (FastAPI): Asynchronous Python service handling file parsing, chunking, and AI orchestration.
3. Database (PostgreSQL 16): Uses pgvector for high-performance vector similarity search (1536 dimensions) alongside relational metadata.
4. AI Core (LangGraph): A state-machine agent that routes intent (Search vs. Chitchat) and enforces citation governance.

• Agentic Routing: Uses LangGraph to classify intent.
• Strict Governance: No hallucinations; every answer includes Citations.
• Enterprise Ingestion: Asynchronous pipeline for PDF/TXT files.
• Modern UX: Responsive React interface with real-time feedback.

• Backend: Python 3.11, FastAPI 0.110, SQLAlchemy Async, Alembic
• AI: LangChain, LangGraph, OpenAI, pgvector
• Frontend: React 18, TypeScript, Tailwind
• Infra: Docker Compose, Nginx

• Docker & Docker Compose
• OpenAI API Key

git clone https://github.com/Nibir1/documind-enterprise.git
cd documind-enterprise
cp .env.example .env
# Add OPENAI_API_KEY
make build

This project includes a comprehensive integration test suite covering 100% of the critical path logic. The tests run inside the Docker container to ensure environment consistency and use AsyncMock to simulate OpenAI and PostgreSQL, ensuring zero-cost, fast execution.

To run the test suite:

make test

• Frontend: http://localhost:3000
• API Docs: http://localhost:8000/docs

• PDF uploaded → Text extracted → 1000-token chunks
• Embedded via text-embedding-3-small → Stored in Postgres

• Router Node → Retriever Node → Generator Node

• Mandatory citations (Filename + Page + Confidence)
• Zero hallucination policy
• Azure Monitor--ready audit logs

• Router decides if query should use search or chitchat

• Recursive chunking & real-time status

documind-enterprise/
├── backend/
│   ├── app/
│   │   ├── api/v1/
│   │   ├── core/
│   │   ├── models/
│   │   ├── schemas/
│   │   └── services/
├── frontend/
│   ├── src/
│   │   ├── api/
│   │   ├── features/
│   │   └── components/
├── docker-compose.yml
└── Makefile

• Core RAG Architecture
• Azure AD (Entra ID) SSO
• Azure Container Apps deployment
• RBAC for document sets

Author: Nahasat Nibir -- Senior Backend Engineer & AI Systems Architect