Regulated sectors
Local authorities, social agencies, central government. Document checks on benefit cases and public procurement, with a chain-of-integrity compatible with legal proceedings.
Use cases
eIDAS-compatible for legal evidence
Full pipeline
Four parallel analysis angles to surface fraud, then four steps to qualify it, document it and handle it.
Spots the pixels that changed.
A dedicated AI model looks at the document at pixel resolution and maps suspect zones, on scans, photos or native PDFs.
Spots the pixels that changed.
A dedicated AI model looks at the document at pixel resolution and maps suspect zones, on scans, photos or native PDFs.
Document coverage
Not just the metadata of native PDFs.
Paper scans, photos, fax.
Handwriting and annotations.
PDFs, e-invoices, signed documents.
Forgery techniques
Targeted edit on a single zone: name, amount, date, IBAN. Often invisible to the naked eye.
e.g. Photoshop, Acrobat, pixel masking
Edits via consumer scan apps and PDF editors. Scan and edit in a single gesture.
e.g. CamScanner, Sejda, PDF24, online tools
AI-generated composites, democratized by the latest generative models. The fastest-growing threat.
e.g. Nano Banana, Midjourney, Stable Diffusion, etc.
Document built entirely from scratch, with no original. Spotted via data inconsistencies and missing history.
e.g. Document built from scratch, fake templates
Patches copy-pasted from one document to another, or zone to zone. Classic technique, losing ground to AI tools.
e.g. Manual copy-paste
Measured performance
From reproducible benchmarks on real corpora, not marketing slides. To be tuned to your infrastructure and your risk tolerance.
01 · Metric
seconds per page
The full pipeline, on a dedicated GPU, processes a page in 1 to 6 seconds depending on resolution and document type. No cloud queue, no quotas.
02 · Metric
automated checks
Twenty-four structural checks on PDFs. Fourteen on JPEG images. Four on PNGs. Each ranked by severity and explained in plain language.
03 · Metric
accuracy on zones
Extraction of sensitive suspect data is measured document by document. Social security number, first name, IBAN: accuracy above 95%. The foundation of reliable qualification.
04 · Metric
of official codes read are verified
Native reading of French 2D-Doc seals and official QR codes. Every readable 2D-Doc seal is cryptographically validated (ECDSA signature against the official list of French authorities). For QR codes pointing to an authentic copy, the document is securely retrieved and compared to the original - any inconsistency on critical fields (name, amount, tax number…) is flagged.
Frequently asked questions
Evaluation
See the application in action on a representative sample of your cases.
