Azure Content Understanding Function

Overview

This Azure Functions app provides the extract_pdf_content HTTP endpoint implemented in function_app.py, which downloads a PDF from Azure Blob Storage and uses Azure AI Content Understanding to extract field data per page, returning the results as JSON.

Key features:

• Managed identity authentication via DefaultAzureCredential.
• Blob retrieval using BlobServiceClient (configurable account/container).
• PDF splitting into single-page byte streams via PyPDF2.
• Parallel page analysis using ThreadPoolExecutor for improved throughput.
• Field extraction per page, with a best-effort TotalTaxesDue value that falls back to TotalDueWithDiscount or TaxAfterDiscount.

Architecture

flowchart LR
    Client(["Client / Caller"])
    AzFunc["Azure Function\nextract_pdf_content\n(HTTP trigger)"]
    Blob[("Azure Blob Storage\nInput PDFs")]
    CU["Azure AI\nContent Understanding"]
    AppInsights["Application Insights\n(Logs & Telemetry)"]

    Client -- "HTTP GET\n?document=<name>.pdf" --> AzFunc
    AzFunc -- "Download PDF blob" --> Blob
    Blob -- "PDF bytes" --> AzFunc
    AzFunc -- "Analyze pages in parallel\n(ThreadPoolExecutor)" --> CU
    CU -- "Extracted fields per page" --> AzFunc
    AzFunc -- "JSON response" --> Client
    AzFunc -. "Logs" .-> AppInsights

Loading

Prerequisites

• Python 3.10+
• Azure Functions Core Tools (for local development)
• Azure Storage account containing the PDFs
• Azure AI Content Understanding endpoint and analyzer (default prebuilt-documentFields)

Configuration

Set the following environment variables (or add them to local.settings.json for local runs):

Setting	Description
BLOB_ACCOUNT_URL	Blob storage endpoint, e.g., https://<account>.blob.core.windows.net.
BLOB_CONTAINER_NAME	Storage container containing PDFs (default documents).
CONTENT_UNDERSTANDING_ENDPOINT	Azure AI Content Understanding endpoint.
CONTENT_UNDERSTANDING_ANALYZER	Analyzer ID (default prebuilt-documentFields).
MANAGED_IDENTITY_CLIENT_ID	Optional user-assigned managed identity client ID for DefaultAzureCredential.
CONTENT_UNDERSTANDING_MAX_CONCURRENCY	Max parallel page analyses (default 4).
TOTAL_TAX_FIELD_CANDIDATES	Comma-separated list of field names treated as "total taxes" (default TotalTaxesDue,TotalDueWithDiscount,TaxAfterDiscount).

Example local.settings.json snippet:

{
	"IsEncrypted": false,
	"Values": {
		"AzureWebJobsStorage": "UseDevelopmentStorage=true",
		"FUNCTIONS_WORKER_RUNTIME": "python",
		"BLOB_ACCOUNT_URL": "https://<account>.blob.core.windows.net",
		"BLOB_CONTAINER_NAME": "documents",
		"CONTENT_UNDERSTANDING_ENDPOINT": "https://<name>.cognitiveservices.azure.com",
		"CONTENT_UNDERSTANDING_ANALYZER": "prebuilt-documentFields",
		"TOTAL_TAX_FIELD_CANDIDATES": "TotalTaxesDue,TotalDueWithDiscount,TaxAfterDiscount"
	}
}

Dependencies

Listed in requirements.txt:

azure-functions
azure-storage-blob
azure-ai-contentunderstanding
azure-identity
PyPDF2

Install locally with:

python -m pip install -r requirements.txt

Running Locally

1. Configure local.settings.json with the required settings.

2. Start the Functions host:

   func start

3. Invoke the endpoint:

   curl "http://localhost:7071/api/extract_pdf_content?document=<blob-name>.pdf"

The response indicates the total number of pages and includes per-page entries with a full fields dictionary and the resolved total tax value.

Deployment Notes

• Deploy using standard Azure Functions workflows (CLI, VS Code, CI/CD).
• Ensure the Function App's managed identity has access to Blob Storage and the Content Understanding resource.
• Tune CONTENT_UNDERSTANDING_MAX_CONCURRENCY according to your hosting plan and analyzer limits.