A bookkeeping client had a year of supplier PDFs in 9 different layouts. The Python pipeline extracts every line, normalizes dates, amounts and vendor names, removes duplicates and checks totals against each invoice, then outputs a CSV for QuickBooks and the dashboard below.
Three of the nine layouts plus a re-sent copy: dates as "03/07/25", "7 Mar 2025" and "2025-03-07"; amounts with and without currency signs; one vendor spelled four ways; a re-sent invoice that would double-count.
| Invoice # | Date | Vendor | Category | Subtotal | Tax | Total | Source |
|---|