Can OCR detect where one invoice ends and another begins?
OCR alone usually cannot do this reliably. The AI layer has to grasp document context, page boundaries, layout, and document type.
What it means
OCR extracts text from a page. It does not, by itself, understand that page 4 ends one invoice and page 5 starts another. untxt. adds a context-aware layer that looks at layout, repeated headers, totals, page numbers, vendor changes, and document type.
02 · Example
A PDF contains invoices with repeated supplier headers and page numbers. untxt. uses layout and content cues to group the right pages together.
Where review matters
When boundaries are unclear, the system should ask for review instead of guessing.
Who this helps
This helps teams that receive combined PDFs from scanners, inboxes, portals, or clients who do not separate documents before uploading them.
What untxt. does
untxt. adds context on top of text extraction. It looks at layout, repeated headers, page numbers, vendor changes, document type, totals, and section breaks to decide which pages belong together.
What it does not pretend
It does not claim OCR alone solves mixed-document intake. Plain OCR can read words; the bookkeeping layer has to understand document boundaries and accounting context.