Extract vendor, totals, and line items from an invoice or receipt with Amazon Textract AnalyzeExpense by parsing SummaryFields and LineItemGroups

Documented steps

1. Grant IAM permission textract:AnalyzeExpense, plus s3:GetObject if using S3Object input.
2. Call AnalyzeExpense with Document={Bytes|S3Object}: https://docs.aws.amazon.com/textract/latest/dg/API_AnalyzeExpense.html — note the AWS CLI does not support the Bytes property, so CLI callers must use S3Object.
3. Enforce sync limits: 10 MB maximum, PNG/JPEG/PDF/TIFF. For larger or multi-page documents use the async StartExpenseAnalysis + GetExpenseAnalysis pair, which follows the same S3-only DocumentLocation pattern.
4. Parse response.ExpenseDocuments (array); each entry has ExpenseIndex, SummaryFields, LineItemGroups, and Blocks.
5. For header data, iterate SummaryFields (ExpenseField objects): read Type.Text (e.g. VENDOR_NAME, INVOICE_RECEIPT_DATE, TOTAL) with Type.Confidence, ValueDetection.Text with its Confidence, and optional LabelDetection.Text for the label printed on the document.
6. For itemized data, iterate LineItemGroups[].LineItems[], each a list of ExpenseField objects (ITEM, QUANTITY, PRICE and similar) with the same Type/ValueDetection/LabelDetection shape.
7. Read the Currency.Code and Currency.Confidence sub-object on monetary fields such as TOTAL for correct currency normalization.
8. Use each ExpenseField's PageNumber to associate fields with the right page on multi-page documents processed through the async path.
9. Handle the documented errors — DocumentTooLargeException, UnsupportedDocumentException, InvalidS3ObjectException, ThrottlingException, InternalServerError — retrying with backoff only on the throttling and 5xx class.
10. Aggregate SummaryFields and LineItemGroups into a normalized invoice object, routing any field with low Type.Confidence or ValueDetection.Confidence to human review.