Mini-project architecture

From frontend screening to a Python verification system

This working Vite/React frontend is the presentation and interaction layer. A production implementation should place sensitive OCR, forensic models, institution matching, and audit logic in a secure Python API.

Implemented in this frontend

Usable now in the browser.

  • Responsive multi-page interface
  • Accessible upload and form validation
  • PDF/JPG/PNG checks up to 10 MB
  • Real SHA-256 file hashing
  • Local duplicate-file detection
  • Rule-based risk explanations
  • Searchable local record history
  • JSON report export and deletion

Requires a Python backend

Not claimed as active in this build.

  • OCR and layout understanding
  • Pixel-level edit and splice detection
  • AI-generated image/text classifiers
  • Signature or seal comparison
  • Authorized issuer database queries
  • Encrypted server-side evidence storage
  • Human-review and audit workflow

Recommended system workflow

A defensible verification result uses several layers rather than one AI score.

01

Document intake

Accept PDF or image files with claimed holder, number, type, issuer, and signature information.

02

Pre-processing

A future Python service can run OCR, normalize images, inspect PDF objects, and extract document structure.

03

Forensic analysis

Combine tamper localization, signature comparison, AI-image detection, typography checks, and duplicate hashing.

04

Authority verification

Connect only to approved institution, school, bank, tax, or identity verification providers with consent.

05

Decision report

Return explainable checks, confidence, risk indicators, and a manual-review path instead of an unsupported claim.

Responsible verification principles

Do not overclaim

A model can flag risk, but “genuine” should require authority confirmation or verifiable cryptographic evidence.

Protect sensitive identity data

Aadhaar, PAN, school, and bank records need consent, masking, encryption, access controls, and a deletion policy.

Keep a human in the loop

High-impact decisions should expose evidence and support manual review, correction, and appeal.

Try frontend screening
VeriDoc AI frontend mini-project · Results are screening indicators, not official authenticity decisions.
Built with GenMB
Built with GenMB