Bank statements into one clean ledger

The situation

Statements arrive monthly as PDFs and CSVs in six different shapes, in fourteen currencies. Merging them by hand is an evening a month, and the moment you re-import a file you either double count or silently drop a genuine duplicate charge.

What we built

• Four statement parsers covering CSV and PDF, including multi-currency card statements with their foreign exchange lines
• Transactions fingerprinted with a parser-specific tiebreaker, so re-importing a file matches perfectly while two identical same-day charges both survive
• A scheduled agent reads bank notification emails and posts previews that the authoritative statement later replaces, keyed on message ID so it stays idempotent
• A rule engine that categorises 99.8 percent of rows, routing the rest to a manual triage screen