TypeScript · Mellow AI

Liedar

Document screening that surfaces evidence of tampering for human review.

What I built

I built a document screening tool that combines file metadata, copied-region detection, and model assessments to surface signs of tampering.

Combined file metadata, image copied-region checks, and model assessments into a shared document-screening pipeline.
Built a browser interface, CLI, and TypeScript library, with structured reports that distinguish tampering, consistency, provenance, and quality signals.
Published a receipt-forgery evaluation and its limitations. The reported historical Jev benchmark does not establish current performance for every provider or document type.

Document screening workflow

PDF / PNG / JPEG
  → read metadata
  → inspect copied image regions
  → assess tampering & consistency
  → return signals for review