AWS announced on Friday (Nov 20) the general availability of Amazon Textract Insights, a new generative AI feature of Amazon Textract, available through a new API. The goal is to end the classic document-processing dilemma: extracting data accurately, but with every value traceable back to its source page — with no model training.
Describe the fields in plain language
With Textract Insights, a team defines what it wants to extract in a blueprint: a JSON Schema in which each field carries a natural-language description of what to find. You can start from AWS public blueprints for common document types — such as W-2, Forms 1040 and 1098, a passport or a birth certificate — or author your own blueprints for your business's documents. Then just submit a single file of up to 3,000 pages from your own Amazon S3 bucket with one asynchronous call (StartDocumentInsights). The service splits the packet into logical sub-documents, matches each one to the best-fitting blueprint and extracts the requested fields.
Confidence and grounding for every value
Every extracted value comes back with a confidence score and its exact location on the page (page and bounding box), so teams can automate high-confidence results and route the rest to human review. There are two kinds of fields: explicit ones, read verbatim from the page, and inferred ones, derived from the document's content — both with a score and traceability. The feature also returns Markdown, HTML and plain text for RAG applications, generates document summaries and figure captions, and detects and redacts personally identifiable information. Symbeo, a company that processes medical bills and forms from many sources and layouts, evaluated the service and highlighted the ability to describe fields in plain language and focus review on the lower-confidence fields.
Textract Core updated, and availability
Alongside the launch, AWS updated Textract's reading engine, Textract Core, with better accuracy, stronger handling of dense tables and handwriting, and expanded coverage to 40 languages in the text detection and analysis APIs — with no changes needed for current customers. Core is also now available in two more regions, Asia Pacific (Tokyo) and Europe (Zurich), for a total of 18 regions. Textract Insights, for now, is available only in US East (N. Virginia) and US West (Oregon). The feature accepts PDF, PNG, JPEG and TIFF, processes documents in English, French, German, Italian, Portuguese and Spanish, and is priced per page, with no token pricing and no minimum commitment.
Image: cover image of the official AWS Machine Learning Blog article announcing Amazon Textract Insights. Image: AWS.
Sources
- AWS Machine Learning Blog — Announcing Amazon Textract Insights: Verifiable generative AI for documents
- AWS — Amazon Textract Insights is now generally available
- Oossa — Amazon Textract Insights launches with AI-powered document extraction
Transparency note: this article was produced with the assistance of artificial intelligence, cross-checked against public posts on X and internet sources, and reviewed before publication. Please consult the original sources to confirm the information.
By Muse