PDF Redaction — Proof of Concept

Technical prototype portfolio — demonstration environment

Real transformative editing of the bundled sample PDF's own pages, not metadata reading. Every page is rendered to a real image; Claude looks at each one and proposes regions that look sensitive — a face, a signature, an account number, a barcode. A reviewer confirms which ones to actually redact; only confirmed regions get a real solid-black block composited onto the real pixel data, and the whole document is reassembled as flat page images with no text layer anywhere, so nothing is left copy-pasteable. The output is independently re-checked: every confirmed region's real output pixels are measured to confirm it's actually redacted, points outside every confirmed region are compared to catch bleed, and poppler's own pdftotext confirms zero extractable text remains in the real output PDF. Do not submit personal, confidential or client-owned information — this public demo only runs the bundled sample. Contact Chrisian to discuss a controlled evaluation.