Extract and Classify Document Data with Azure AI

This skill teaches an assistant to use the Azure AI Document Intelligence SDK for .NET to extract text, tables, and structured data from documents using prebuilt and custom models. The current version is v1.0.0 (GA). It covers authentication, the two client types, the prebuilt models, seven core workflows, and error handling.

Two clients do the work. DocumentIntelligenceClient analyzes and classifies documents. DocumentIntelligenceAdministrationClient builds and manages custom models and classifiers. The prebuilt models are prebuilt-read (text, languages, handwriting), prebuilt-layout (text, tables, selection marks, structure), prebuilt-invoice (vendor, items, totals), prebuilt-receipt (merchant, items, total), prebuilt-idDocument (name, date of birth, address), prebuilt-businessCard, prebuilt-tax.us.w2, and prebuilt-healthInsuranceCard.us.

The seven core workflows are analyzing an invoice, extracting layout (text by page, lines, words, tables and cells), analyzing a receipt, building a custom model, building a document classifier, classifying a document, and managing models (resource details with the custom model count against its limit, getting, listing, and deleting models). For custom models there are two build modes: Template for fixed layout documents such as forms, and Neural for variable layout documents.

Use it when you are writing C# code that must read invoices, receipts, IDs, business cards, tax forms, or general documents, or when you need to train a custom extraction model or a classifier over labelled training data in blob storage.

Do not use it for a task that does not clearly match this scope. Do not treat the output as a substitute for environment-specific validation, testing, or expert review, and stop and ask for clarification if required inputs, permissions, safety boundaries, or success criteria are missing. The older Azure.AI.FormRecognizer package is described as a deprecated legacy SDK, so use DocumentIntelligence instead. The skill is marked with a critical risk level in its metadata.

Inputs are a document URI, the model ID to use, and configuration through three environment variables: DOCUMENT_INTELLIGENCE_ENDPOINT, DOCUMENT_INTELLIGENCE_API_KEY, and BLOB_CONTAINER_SAS_URL for training data. Authentication can use Microsoft Entra ID, which is recommended and requires a custom subdomain rather than a regional endpoint, or an API key.

Outputs are an AnalyzeResult holding analyzed documents, pages, and tables. Each extracted DocumentField carries a value and a confidence score. Install the package with:

dotnet add package Azure.AI.DocumentIntelligence

• Azure.Identity, whose DefaultAzureCredential handles Entra ID authentication.
• Azure Blob Storage, which supplies training data to custom models and classifiers through a SAS URL and BlobContentSource, optionally with a prefix per document type.
• Document Intelligence Studio and the NuGet package, API reference, and GitHub samples listed in the skill.

Best practices: use DefaultAzureCredential in production, reuse client instances because they are thread-safe, use WaitUntil.Completed for long-running operations for simplicity, always check the Confidence property on fields, choose prebuilt models for common documents and custom ones for specialized cases, and use a custom subdomain when authenticating with Entra ID. Failures surface as RequestFailedException, which exposes a status and message.

It is for .NET developers building document processing, data entry automation, or intake pipelines on Azure, and for teams that need custom document models or classifiers trained on their own labelled samples.