The receipt pipeline combines deterministic parsing and model-assisted scoring. This gives explainability plus adaptive fraud resistance.

Upload intake and file-type checks
OCR extraction (date, amount, station, wallet chars)
Image integrity analysis (tamper/AI likelihood)
Perceptual hash generation for duplicate detection
Gate outputs into policy decision engine

  RECEIPT INTELLIGENCE PIPELINE
  ─────────────────────────────

  ┌────────────┐   ┌────────────┐   ┌────────────┐   ┌────────────┐   ┌────────────┐
  │  UPLOAD    │ → │  OCR PARSE │ → │  IMAGE     │ → │  DUPLICATE │ → │  GATE      │
  │  INTAKE    │   │            │   │  INTEGRITY │   │  CHECK     │   │  INPUTS    │
  │            │   │  Extract:  │   │            │   │            │   │            │
  │  File type │   │  · Date    │   │  AI score  │   │  SHA-256   │   │  All scores│
  │  File size │   │  · Amount  │   │  Tamper    │   │  pHash     │   │  ready for │
  │  Format OK │   │  · Station │   │  score     │   │  compare   │   │  policy    │
  │            │   │  · Wallet  │   │            │   │  vs all    │   │  engine    │
  └────────────┘   └────────────┘   └────────────┘   └────────────┘   └────────────┘

  Outputs saved: OCR confidence · AI/tamper scores · duplicate flags · receipt metadata

Receipt Intelligence Pipeline