AIFondry-ContentUnderstanding

🎓 Demo Scripts - Content Understanding Walkthrough

Overview

This folder contains a progressive 3-step walkthrough for Azure AI Foundry Content Understanding:

1. Prebuilt analyzers for zero-configuration extraction from standard business documents
2. Custom extraction for domain-specific entities, summaries, and document classification
3. A multi-document pipeline that classifies, routes, and consolidates results across mixed inputs

Together, the demos move from simple document analysis to a production-style routing pipeline.

Prerequisites

• Python 3.9+
• An Azure AI Foundry resource with Content Understanding enabled
• A configured .env file in the repository root
• Network access to the sample documents hosted in the Azure samples repository

Your repo root .env must include:

CONTENTUNDERSTANDING_ENDPOINT=https://<your-resource>.cognitiveservices.azure.com/
CONTENTUNDERSTANDING_KEY=<your-key>

Setup

From the repository root:

python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r demos\requirements.txt

Then make sure .env exists in the repo root. If needed, copy the template first:

Copy-Item .env.template .env

Update .env with:

• CONTENTUNDERSTANDING_ENDPOINT
• CONTENTUNDERSTANDING_KEY

Script Descriptions

01_prebuilt_analyzer.py

What it does

• Uses the built-in prebuilt-invoice and prebuilt-receipt analyzers
• Analyzes sample invoice and receipt documents from public URLs
• Prints markdown previews plus extracted structured fields

Key concepts demonstrated

• Using ContentUnderstandingClient
• Calling begin_analyze() with public document URLs
• Working with prebuilt analyzers
• Interpreting field confidence scores
• Rendering nested field output in tables

How to run it

python demos\01_prebuilt_analyzer.py

Expected output

• Rich console panels for the invoice and receipt runs
• A markdown preview for each document
• A table of extracted fields such as totals, vendor details, dates, and line-item-related values
• Color-coded confidence values for quick triage

02_custom_entity_extractor.py

What it does

• Creates a temporary custom analyzer on top of prebuilt-document
• Extracts entities such as people, organizations, dates, monetary amounts, and locations
• Generates a one-paragraph summary, key terms, and a classified document type
• Analyzes a sample invoice, then deletes the analyzer during cleanup

Key concepts demonstrated

• Defining a custom ContentFieldSchema
• Using all three generation methods: extract, generate, and classify
• Enabling OCR, layout, and confidence/source estimation
• Creating and deleting analyzers programmatically

How to run it

python demos\02_custom_entity_extractor.py

Expected output

• A confirmation that a timestamped custom analyzer was created
• A table listing configured fields and methods
• Panels showing the detected document type, generated summary, extracted entities, and key terms
• Cleanup messages confirming the analyzer was deleted

03_multi_document_pipeline.py

What it does

• Creates two temporary analyzers:
  • an entity/summary analyzer for general documents
  • a classifier that routes documents to the correct analyzer
• Processes multiple documents in sequence
• Produces console summaries plus a consolidated JSON report
• Deletes the analyzers during cleanup

Key concepts demonstrated

• Document classification and analyzer routing
• Batch processing with progress reporting
• Mixing prebuilt analyzers with custom analyzers
• Normalizing results into a single JSON structure
• Building a production-style pipeline for mixed document sets

How to run it

python demos\03_multi_document_pipeline.py

Expected output

• Progress bars while documents are processed
• Per-document routing summaries showing category and analyzer used
• A classification summary table for all processed documents
• A saved analysis_report.json file containing the consolidated results
• Cleanup messages for both temporary analyzers

Key Concepts

Method	Purpose	Example in demos
extract	Pull literal values directly from the document content	People, organizations, dates, monetary amounts, and locations in 02_custom_entity_extractor.py
generate	Produce AI-generated content derived from the source document	summary and key_terms in 02_custom_entity_extractor.py, plus summary in 03_multi_document_pipeline.py
classify	Assign the document to a category for routing or labeling	document_type in 02_custom_entity_extractor.py, and category-based routing in 03_multi_document_pipeline.py

Confidence Scores

Use confidence scores as a lightweight decision policy:

Confidence	Suggested action
> 0.85	Auto-approve
0.5 - 0.85	Send for human review
< 0.5	Reject or reprocess

This matches the color-coded thresholds used in the prebuilt analyzer demo.

Troubleshooting

• Missing .env
  Copy the template from the repo root:

  Copy-Item .env.template .env
  

• Invalid endpoint
  Verify CONTENTUNDERSTANDING_ENDPOINT in the Azure Portal and ensure it matches your Azure AI Foundry resource endpoint.

• Model not deployed
  Open your Content Understanding resource settings and confirm the required models and analyzer capabilities are available.

• Rate limiting
  If you see throttling or transient API failures, add delays between requests or reduce batch size while testing.

Resources

• Content Understanding overview
• Python SDK
• SDK samples
• Custom analyzer tutorial

See Also

• Main README - Project overview, prerequisites, and full documentation links
• Quickstart Demo - Run your first demo in 2 minutes
• Infrastructure (Bicep / Terraform) - Deploy the Azure resource
• Model Explanation - Why each AI model is needed

This site is open source. Improve this page.