Document Intelligence Platform
Enterprise-grade document processing platform with OCR, graph-based workflow orchestration, document classification, custom model training, and human-in-the-loop review capabilities.
Platform Overview
The Document Intelligence Platform is a comprehensive solution for document processing workflows. It combines OCR services, flexible graph-based workflow orchestration, supervised learning capabilities, and collaborative review tools into a unified platform.
🚀 Core Features
- Document Processing - Multi-format document upload and OCR with Azure Document Intelligence
- Document Classification - Train and deploy Azure Document Intelligence classifiers for automated document type routing
- Graph Workflows - DAG-based workflow engine with 30+ extensible activities powered by Temporal
- Model Training - Supervised learning with custom model training and versioned deployments
- HITL Review - Collaborative human-in-the-loop workflows with confidence-based routing
🏗️ Architecture
- Backend - NestJS with Express, Prisma ORM, PostgreSQL
- Frontend - React with TypeScript, Mantine UI
- Orchestration - Temporal workflow engine with durable execution
- Storage - Pluggable blob storage (Azure, S3, filesystem)
Key Capabilities
Graph-Based Workflow Engine
The platform features a sophisticated DAG (Directed Acyclic Graph) workflow engine that enables complex, multi-step document processing pipelines. Workflows are defined as JSON graphs with nodes representing activities like OCR, validation, training, and review operations.
30+ Built-in Activities: OCR processing, field extraction, validation, confidence filtering, model training, HITL routing, parallel execution, conditional branching, and more.
📄 Document Management
Full lifecycle document management with metadata tagging, versioning, and relationship tracking. Supports PDF, images, and multi-page documents.
🏷️ Labeling & Annotation
Project-based labeling system for ground truth creation. Supports bounding box annotations, field labeling, and batch operations for training data preparation.
🧠 Model Training
End-to-end custom model training pipeline. Train Azure Document Intelligence custom models from labeled data with versioning and deployment management.
👥 Human Review Queue
Collaborative HITL system with assignment management, bulk operations, and confidence-based routing for quality assurance workflows.
🔐 Authentication & Security
Keycloak SSO integration with JWT bearer tokens and API key management. Role-based access control for secure multi-tenant operations.
📊 Extensible Activities
Developer-friendly activity registry system. Create custom activities by extending base classes and registering them in the workflow engine.
🗂️ Document Classification
Train Azure Document Intelligence classifiers to automatically identify document types. Manage the full lifecycle from training data upload through classifier deployment.
Quick Start
Node.js 24+, PostgreSQL 14+, Temporal Server, Azure Document Intelligence resource (for OCR)
1. Install Dependencies
npm install
2. Configure Environment
Set up environment variables for database, Temporal, Azure services, and authentication:
DATABASE_URL=postgresql://user:pass@localhost:5432/docdb
TEMPORAL_ADDRESS=localhost:7233
AZURE_DI_ENDPOINT=https://your-instance.cognitiveservices.azure.com/
AZURE_DI_KEY=your-key
SSO_AUTH_SERVER_URL=https://your-keycloak.com/realms/your-realm
3. Initialize Database
cd apps/backend-services
npm run db:migrate
npm run db:generate
4. Start Services
# Terminal 1: Backend API
cd apps/backend-services && npm run start:dev
# Terminal 2: Temporal Worker
cd apps/temporal && npm run worker
# Terminal 3: Frontend
cd apps/frontend && npm run dev
5. Access the Platform
- Frontend: http://localhost:3000
- Backend API: http://localhost:3002
- API Docs: http://localhost:3002/api
- Temporal UI: http://localhost:8233
API Documentation
Comprehensive REST API documentation is available on the API Reference page. The API supports both Bearer token authentication (Keycloak SSO) and API key authentication.
📚 API Reference
Interactive API documentation with request/response schemas, authentication details, and example requests for all 56 endpoints.
💻 GitHub Repository
View source code, report issues, contribute to the project, and access additional documentation.
Technology Stack
Core Technologies
Backend Services
- NestJS 11 + Express
- Prisma ORM
- PostgreSQL 14+
- Temporal Workflow Engine
- Azure Document Intelligence SDK
Frontend Application
- React 19 + TypeScript
- Vite
- Mantine UI Components
- React Flow (workflow visualization)
- Axios + React Query
Project Structure
apps/
├── backend-services/ # NestJS API server (14 modules)
├── frontend/ # React application
├── temporal/ # Temporal workflow workers
├── image-service/ # Python image preprocessing tools
└── shared/ # Shared Prisma schema
docs-md/ # Technical documentation
docs/ # This documentation site
deployments/ # OpenShift/Kubernetes manifests
Learn More
Detailed technical documentation is available in the
/docs-md directory including:
- HITL_ARCHITECTURE.md: Human-in-the-loop system design
- TEMPLATE_TRAINING.md: Model training workflows
- docs-md/graph-workflows/: Workflow engine architecture and activity development guides