Where the product is today

  • Invoice review supports explicit page-range splitting.
  • Automatic general document classification and mixed-packet splitting are not available.

An invoice and delivery note in one PDF

The planned workflow would separate the records, preserve their original pages and apply the appropriate extraction schema to each.

Availability and evaluation

  • The current product does not automatically classify arbitrary uploads.
  • A page break or repeated title alone does not prove a document boundary.

What we plan to build

  • Classify document type and route to a versioned extraction schema.
  • Detect multi-document boundaries and retain the original page mapping.
  • Use confidence thresholds and an unknown class for uncertain cases.
  • Let reviewers merge, split and correct document groups before downstream export.

How this should be evaluated

  • Measure both classification and boundary accuracy.
  • Include blank pages, cover sheets and unknown document types.
  • Confirm that every original page belongs to a reviewed group or an explicit exception.