Demo build / AI automation
Invoice ExtractorInvoices in, clean data out
Invoices and receipts go in; validated, structured data comes out — ready for a spreadsheet or accounting tool, with anything uncertain flagged for a human.
- Type
- AI automation · demo build
- My role
- Solo build — product, design and engineering
- Stack
- Next.js · Claude API · Structured outputs · n8n
01The challenge
Finance teams still retype invoices and receipts into spreadsheets: slow, boring and error-prone, especially with mixed currencies, discounts, VAT and crumpled thermal receipts. This demo uses fictional sample documents to show that work done automatically — and checked.
02What I built
- AI extraction of vendor, dates, line items, tax and totals against a strict schema.
- Automatic checks that re-add the numbers, with anything that doesn't reconcile flagged for review.
- Editable results with JSON view and CSV export, ready for a sheet or accounting tool.
- The same pipeline as an n8n workflow: invoices emailed to an inbox land in Google Sheets, with a Slack summary for approval.
03Key decisions
The technical calls that mattered — and what they meant for the business.
Validated, not trusted blindly
Output must match a strict schema and the totals are re-checked in code. Unclear fields are marked for a human instead of slipping silently into the books.
Works where the team already works
Shipping it as an n8n workflow means invoices flow from Gmail into Google Sheets and Slack — no new app for the finance team to learn.
Built for messy real-world documents
Samples include a Nigerian supplier invoice with discount and VAT, and a crooked GBP café receipt — not just clean PDFs.
A demo that costs nothing to host
Visitors pick a sample and see output saved from a real run of the pipeline, so the public site makes no AI calls.