DataVance

DVEX Receipt & Expense Extraction

Receipts are the worst documents in accounting

Crumpled thermal paper, faded ink, a photo taken at an angle on a loading dock. DVEX reads them anyway — merchant, ABN, GST, totals, payment method, and an expense category, each field carrying its own confidence score so your pipeline knows what to trust and what to queue for a human.

Live on RapidAPI. Subscribe on the marketplace — authentication and billing are handled there, so a key is all you need.

POST/api/v1/receipts/process200 OK
{
  "extracted": {
    "merchant_name": { "value": "Bunnings", "confidence": 0.98 },
    "abn": { "value": "51824753556", "confidence": 0.97 },
    "receipt_date": { "value": "2026-06-28", "confidence": 0.97 },
    "gst": { "value": 4.39, "confidence": 0.96 },
    "total": { "value": 48.29, "confidence": 0.98 },
    "card_last4": { "value": "1234", "confidence": 0.90 },
    "category": { "value": "hardware", "confidence": 0.60 }
  },
  "extraction_confidence": 0.60,
  "validation_errors": []
}

What comes back, and what it means

extraction_confidence
The lowest confidence across every field that was found. One number to route on: above your threshold, post it; below, queue it.
category
One of fifteen expense categories — groceries, fuel, travel, accommodation, software, hardware and so on — ready for GL coding.
gst
Checked against the Australian rule, not just read: GST should be one eleventh of a GST-inclusive total.
card_last4
Only ever the last four digits. If a receipt prints a full card number, the API reduces it before responding — enforced in code on every extraction path.
validation_errors
Named problems rather than silent guesses: a failed ABN checksum, GST that doesn't reconcile, line items that don't sum to the total.

The same engine, already benchmarked

DVEX runs on the extraction engine behind DVAP, whose accuracy is measured against public datasets — including SROIE, a receipt dataset — and published openly rather than claimed. The engine routes each document the cheapest capable way: text is read directly from digital PDFs at no OCR cost, while photos and scans go to the vision model.

Nothing is stored

Receipts are processed in memory and discarded. There is no document store to breach and no retention window to explain, because the API keeps nothing you send it.

One receipt, one request

Deliberately no batch endpoint. Marketplace billing counts requests, so one request equals one receipt — what you are charged is exactly what you processed.

What DVEX guarantees

Every line below is checkable rather than asserted — that is the whole point of the brand.

  • Endpoint deployed and answering in production
  • 15 automated tests, part of a 150-test suite
  • Card-number reduction verified on every extraction path
  • OpenAPI specification written and validated

Start using DVEX

Plans, quotas, and rate limits are on the listing itself, so what you read there is always what you are billed — we never restate prices here where they would go stale. There is a free tier to try it on your own documents first. Processing supplier invoices as well? DVAP extracts, validates, fraud-checks, and PO-matches them on the same engine, with the accuracy published.