A receipt scanner is a genuinely good student project: it's visual, it has a clear success metric, and it teaches the exact pattern used in real document-AI products.

The pipeline

  1. Photograph or upload the receipt image.
  2. Send it to a vision-capable model with a prompt asking for a strict JSON schema: merchant, date, line items, total.
  3. Validate the response against the schema and re-prompt once if it fails to parse.
  4. Cross-check: do the line items sum to the stated total, within a small tolerance for rounding?

The check that matters most

Step 4 is the whole project's credibility. A model can extract plausible-looking numbers that don't actually add up; catching that automatically, and flagging the receipt for manual review when they don't, is what separates a toy demo from something you'd trust with real expense data.

See document AI for invoices: an extraction schema for the schema-design half of this problem.

— Pranjul Rathour, GenAI Engineer from Kanpur, India. Open to GenAI roles, hackathon judging, mentorship sessions and guest talks at any campus: pranjulrathour41@gmail.com.