AI Document Classification and Extraction

---

⚠️ IMPORTANT NOTE

The code generated by LLMs may contain errors or not follow all best practices. Always thoroughly test the generated code in your development environment and perform code reviews before deploying to production.

---

A demo and guide for setting up and using Document Intelligence for custom classification, with integration into content understanding and other workflows. This beginner friendly repository provides:

• Step-by-step guide to implementing a Document Intelligence Classification use case using Microsoft Azure.
• Integration for further extraction using Content Understanding in Microsoft Azure Foundry.
• Model evaluation using custom notebook to generate accuracy metrics.

---

Repository Structure

1. Microsoft Azure Document Intelligence Model Guide

Location: Contains:

2. Document Intelligence Studio Guide

Location: Contains:

3. Custom Notebook options

Location: Contains:

Features

• Multi-Environment Support: Seamless integration with Azure AI Foundry and Document Intelligence Studio for classification workflows.
• Custom Classification Models: Train and deploy models to classify documents by type using labeled folders and Azure Blob Storage.
• No-Code Studio Interface: Use Document Intelligence Studio to build, train, and test classifiers without writing code.
• Content Understanding Integration: Chain classification output into Azure AI Foundry’s Content Understanding for schema-based extraction and downstream processing.
• Scalable Architecture: Supports large-scale document ingestion and classification with confidence scoring and schema validation.
• Extensible Framework:
  • 🔄 Adaptable for various document types and formats
  • 🎯 Supports custom folder structures and labeling strategies
  • 🧩 Modular design for integrating with downstream AI services
  • 📚 Reusable setup for multiple classification use cases

Getting Started

1. Environment Setup:

• Install the https://learn.microsoft.com/en-us/cli/azure/install-azure-cli and verify with az --version
• Log in using az loginmca
• Create a resource group in Azure Foundry
• Provision Azure Resources:
  • Create a Document Intelligence resource in the Azure Portal
  • Note the key and endpoint for API access
• Set up a Storage Account and Blob Container for training data
• Upload Training Data:
  • Organize documents into folders by class
  • Use az storage blob upload-batch to upload folders to your container

2. Train a Classifier:

• Open Document Intelligence Studio
• Create a new custom classifier project
• Select the container and folder structure
• Run training and validate model performance
• Integrate with Content Understanding:
  • Use the classifier output as input to Azure AI Foundry’s Content Understanding
  • Define schemas for structured extraction
  • Deploy as part of a scalable document processing pipeline 4

Requirements

• Python 3.x (for optional automation)
• Azure CLI
• Azure subscription with access to:
  • Azure AI Document Intelligence
  • Azure AI Foundry
  • Storage Account and Blob Container
  • Labeled training data (minimum 5 documents per class)

License

See the LICENSE file for details.