Extract and Classify Document Data with Azure AI

Azure.AI.DocumentIntelligence is a .NET SDK for extracting text, tables, and structured data from documents using either Microsoft's prebuilt models or custom-trained ones. Two client types cover the whole surface: DocumentIntelligenceClient for analyzing and classifying documents, and DocumentIntelligenceAdministrationClient for building and managing custom models and classifiers. Eight named prebuilt models cover common document types out of the box - prebuilt-read (text, languages, handwriting), prebuilt-layout (text, tables, selection marks, structure), prebuilt-invoice, prebuilt-receipt, prebuilt-idDocument, prebuilt-businessCard, prebuilt-tax.us.w2, and prebuilt-healthInsuranceCard.us. Installation is a two-package add:

dotnet add package Azure.AI.DocumentIntelligence
dotnet add package Azure.Identity

Use this SDK when a .NET application needs to pull structured fields out of documents - vendor name and line items from an invoice, merchant and total from a receipt, or arbitrary text, tables, and selection marks via the layout model - rather than parsing raw document bytes by hand. When prebuilt models don't fit a specialized document type, build a custom model (DocumentBuildMode.Template for fixed-layout forms, DocumentBuildMode.Neural for variable layouts) trained from a blob-container source, or a document classifier to route documents by type before analysis. It supersedes Azure.AI.FormRecognizer, which is explicitly called out as the legacy, deprecated SDK - new projects should use DocumentIntelligence instead. It is not the right tool if the task doesn't involve document field or structure extraction at all.

Input is a document URI, or a blob-container source for training data; output is an AnalyzeResult containing AnalyzedDocument entries with named DocumentFields typed as string, date, number, currency, or list, each carrying a Confidence score, plus page-level DocumentPage/DocumentLine/DocumentTable data for the layout model. For custom models, output is DocumentModelDetails with a per-document-type field schema and confidence; for classifiers, output is a document type label with its own confidence score per analyzed document.

Authenticates via Microsoft Entra ID (DefaultAzureCredential, recommended for production, requiring a custom subdomain endpoint rather than a regional one) or a plain AzureKeyCredential API key; training data for custom models and classifiers comes from Azure Blob Storage via a SAS-URL-scoped BlobContentSource. Errors surface as RequestFailedException with a status code and message. Client instances are thread-safe and meant to be reused rather than recreated per call, long-running operations are typically awaited with WaitUntil.Completed for simplicity, and every extracted field's Confidence property should be checked before trusting its value.

.NET developers extracting structured data from invoices, receipts, IDs, business cards, tax forms, or custom document types, who want typed, confidence-scored fields from prebuilt or custom Azure models instead of hand-rolled document parsing.