What is Azure AI Document Intelligence

In this comprehensive article, I’ll walk you through everything you need to know about Azure AI Document Intelligence, from its core functionalities to advanced implementation strategies.

What is Azure AI Document Intelligence

Azure AI Document Intelligence is a cloud-based service powered by Microsoft Azure that uses advanced machine learning models to extract, analyze, and understand content from documents. It goes beyond basic Optical Character Recognition (OCR) by not just recognizing text but comprehending the structure, context, and relationships within documents.

At its core, Document Intelligence analyzes your forms and documents, extracts text and data, and maps field relationships as key-value pairs. It returns a structured JSON output that’s ready for further processing or integration with other systems.

Key Capabilities

Here are the primary capabilities

• Advanced OCR: Precisely extracts text from various document types, including scanned images
• Layout Analysis: Understands document structure, including tables, headings, and sections
• Key-Value Pair Extraction: Automatically identifies fields and their values
• Table Recognition: Accurately extracts tabular data with relationship preservation
• Form Recognition: Processes structured forms like invoices, receipts, and ID cards
• Custom Models: Trains specialized models for your unique document types

How Azure AI Document Intelligence Works

The technology behind Azure AI Document Intelligence combines computer vision and natural language processing to transform unstructured document content into structured, actionable data. Here’s a simplified view of the process:

1. Document Ingestion: The service accepts various document formats, including PDFs, JPEGs, PNGs, and TIFFs
2. Layout Analysis: Advanced algorithms identify document structure, text, tables, and other elements
3. Text Extraction: OCR technology converts visual text to a machine-readable format
4. Semantic Understanding: AI models interpret the meaning and context of extracted text
5. Data Structuring: Information is organized into a structured JSON format ready for integration

Types of Models

One of the most powerful aspects of Document Intelligence is its versatility through different model types. Let’s explore the main categories:

Pre-built Models

Microsoft has done the heavy lifting by creating specialized models for common document types:

Custom Models

While pre-built models are well-suited for standard documents, custom models excel when handling industry-specific or proprietary documents. I’ve helped clients develop custom models for:

• Insurance claim forms
• Medical records
• Legal contracts
• Financial statements
• Logistics documentation

Setting Up Azure AI Document Intelligence: A Step-by-Step Tutorial

Follow the steps below.

1. Create an Azure AI Document Intelligence Resource

First, you’ll need an Azure account. If you don’t have one, you can sign up for a free trial.

1. Log in to the Azure portal.
2. Click “Create a resource”
3. Search for “Document Intelligence” and select it
4. Fill in the required details:
   • Subscription: Your Azure subscription
   • Resource group: Create new or use existing
   • Region: Select a supported region (East US or West US for US-based operations)
   • Name: Give your resource a unique name
   • Pricing tier: Select based on your volume needs (Free tier available for testing)
5. Click “Review + create” and then “Create”. Refer to the screenshot below for better clarity.

2. Gather Your API Credentials

Once your resource is created:

1. Navigate to your Document Intelligence resource
2. Go to the “Keys and Endpoint” section
3. Note down the following:
   • Endpoint URL
   • API Key (you’ll have two keys; either one works)

These credentials will be essential for authenticating your API requests. Refer to the screenshot below for complete details.

3. Choose Your Implementation Method

Azure AI Document Intelligence offers multiple ways to interact with the service:

REST API

For direct integration with any programming language, the REST API provides the most flexibility. Here’s a simple example using Python:

import requests
import json

endpoint = "YOUR_ENDPOINT"
api_key = "YOUR_API_KEY"
model_id = "prebuilt-layout"  # Using the layout model

# Prepare the request headers
headers = {
    "Content-Type": "application/json",
    "Ocp-Apim-Subscription-Key": api_key
}

# Document URL (you can also upload a document directly)
document_url = {"source": "https://example.com/sample-document.pdf"}

# Make the API call
response = requests.post(
    f"{endpoint}/documentintelligence/documentModels/{model_id}:analyze?api-version=2023-07-31",
    headers=headers,
    json=document_url
)

# Process the response
results = response.json()
print(json.dumps(results, indent=2))

Client Library

For more streamlined development, I recommend using the client libraries available for languages like Python, .NET, Java, and JavaScript. Here’s a .NET example:

using Azure;
using Azure.AI.DocumentIntelligence;
using Azure.AI.DocumentIntelligence.Models;

// Set up the client
string endpoint = "YOUR_ENDPOINT";
string apiKey = "YOUR_API_KEY";
AzureKeyCredential credential = new AzureKeyCredential(apiKey);
DocumentIntelligenceClient client = new DocumentIntelligenceClient(new Uri(endpoint), credential);

// Analyze the document using a prebuilt model
using var stream = File.OpenRead("sample-document.pdf");
AnalyzeDocumentOperation operation = await client.AnalyzeDocumentAsync(WaitUntil.Completed, "prebuilt-layout", stream);
AnalyzeResult result = operation.Value;

// Process extracted content
foreach (DocumentPage page in result.Pages)
{
    Console.WriteLine($"Page {page.PageNumber} has {page.Lines.Count} lines.");
    foreach (DocumentLine line in page.Lines)
    {
        Console.WriteLine($"Line: '{line.Content}'");
    }
}

Studio Interface

For non-developers or quick testing, the Document Intelligence Studio offers a user-friendly web interface that allows you to analyze documents without writing code.

Best Practices

Here are some best practices to follow

1. Document Preparation

• Ensure good image quality (300 DPI or higher)
• Correct any skew or orientation issues
• Use OCR-friendly fonts when possible
• Remove any unnecessary markings or annotations

2. Model Selection

• Start with pre-built models for standard document types
• Use the Layout model for general document understanding
• Invest in custom models for domain-specific documents
• Consider composite models for complex workflows

3. Testing and Validation

• Always validate model outputs against ground truth
• Test with a diverse set of document examples
• Pay special attention to edge cases
• Implement human review for low-confidence extractions

4. Integration Considerations

• Use Azure Logic Apps or Power Automate for workflow integration
• Consider Azure Functions for serverless processing
• Implement error handling and retry mechanisms
• Set up monitoring and logging for production deployments

Conclusion

Azure AI Document Intelligence is one of the most powerful document processing platforms available today. Its combination of pre-built intelligence and customization options makes it suitable for a wide range of document processing challenges.

For organizations that wish to automate manual data entry or extract insights from document archives, Document Intelligence offers a better solution.

With the right approach, Azure AI Document Intelligence can transform your document processing.

You may also like the following articles.

• What Are the Main Features of Azure AI Foundry
• List of Azure AI Services
• What is Azure AI services multi-service account

Rajkishore

I am Rajkishore, and I am a Microsoft Certified IT Consultant. I have over 14 years of experience in Microsoft Azure and AWS, with good experience in Azure Functions, Storage, Virtual Machines, Logic Apps, PowerShell Commands, CLI Commands, Machine Learning, AI, Azure Cognitive Services, DevOps, etc. Not only that, I do have good real-time experience in designing and developing cloud-native data integrations on Azure or AWS, etc. I hope you will learn from these practical Azure tutorials. Read more.