Bank statement parser: from PDF to structured data

Lenders and accounting firms receive PDF bank statements from clients. Manually entering transactions takes time and invites mistakes, while external services charge for every document.

We built a service that extracts the account holder, account number, period, balances, and full transaction list from a PDF. Every result passes a balance check: the opening balance plus all transactions must equal the closing balance. A separate layer checks whether the document was altered. The service works as an API with an access key and as a dashboard for manually checking files; files are not saved to disk.

Python, FastAPI, Pydantic v2, custom parsers for each bank, tests on real documents.