From the source
Amazon made Amazon Textract Insights generally available, a generative AI feature of Amazon Textract that lets teams describe the fields they want in natural language and get structured, auditable results.
- The feature is reached through a new API inside Amazon Textract, and Amazon says no model training is required.
- Users define fields in a blueprint, a JSON Schema where each field carries a natural language description and an inference type that decides whether a value is read verbatim or inferred.
- For mixed packets, blueprints for each expected document type go into one project, and a single file of up to 3,000 pages is submitted from an Amazon S3 bucket with splitting enabled.
- Amazon says the service splits the packet, matches each sub-document to a blueprint, extracts the fields, and returns a confidence score plus page-level grounding for every value.
- Symbeo's Vijay Agrawal said the confidence scores let the team focus review on lower-confidence fields, and that the company is excited to explore document splitting.
- Amazon also updated the Textract reading engine, Textract Core, improving accuracy, dense tables and handwriting, and expanding language coverage to 40 languages for the DetectDocumentText and AnalyzeDocument APIs.






