AWS has announced general availability of Amazon Textract Insights, a generative AI feature in Amazon Textract exposed through its own new API, according to the AWS Machine Learning Blog.
What it does
You define what to extract in a blueprint, a JSON Schema where each field has a plain-language description. Each field also has an inference type. It says whether the value is read word for word from the page or inferred from what the document contains. AWS provides public blueprints for common documents such as a W-2, Form 1040, Form 1098, a passport and a birth certificate. You may draft your own versions for paperwork unique to your company. If you're handling a mixed bundle, include the templates for each document category you anticipate within a single project, then upload one file of as many as 3,000 pages from your personal Amazon S3 storage location, with splitting turned on. The service divides the file into logical sub-documents, matches each to a blueprint, extracts the fields and writes results back to your bucket. AWS gives the example of a 26-page mortgage application holding a W-2, payslips, bank statements and handwritten loan officer notes, processed with one StartDocumentInsights call.
Confidence and traceability
Every value comes back with a confidence score and grounding: the page and the bounding box it came from. That applies to both explicit fields and inferred ones. A reviewer or auditor can check any result against the original page. AWS positions this as a way to automate documents that used to go to manual review while keeping people in the loop where judgment is needed.
Symbeo, a CorVel company, evaluated the service for medical bills and forms that arrive in many layouts. Its information systems vice president said the team could describe fields in plain language and focus review on low-confidence values. Symbeo is also interested in the splitting feature, which it has not yet put into use.
Core engine update
AWS also updated the Amazon Textract Core reading engine. It improves accuracy, dense tables and handwriting, and covers 40 languages for DetectDocumentText and AnalyzeDocument. Existing APIs are unchanged. Core is now available in Asia Pacific (Tokyo) and Europe (Zurich), which brings in-region processing to 18 AWS Regions. Insights is not yet available in those two Regions.
Data handling
Projects and blueprints are encrypted at rest with AWS owned keys, or with your own KMS customer managed key if you set one at creation. Intermediate job data is transient and deleted when the job completes. Insights uses cross-Region inference, so requests are routed to compute within the geography where they originated. For example, requests from the United States stay in the United States. Core APIs keep processing in the Region you call.
Limits and cost
Inputs can be PDF, PNG, JPEG or TIFF. Insights processes English, French, German, Italian, Portuguese and Spanish. Pricing is per page rather than per token, with no minimum commitment. AWS advises tuning the acceptance threshold on a representative sample of your own documents.