Banking / compliance · Veganext · Jun-Jul 202696.1% extraction accuracy
On-premise document verification & field extraction
On-premise service for a Tunisian bank's securities-order and fund-subscription forms. Given a scanned PDF, it verifies the four mandatory marks - agency stamp, bank signature, "signature conforme," client signature - extracts around 30 structured fields, and returns a verdict with an explainable 0-100 trust score.
- 96.1% extraction accuracy on the demo set and 91.3% on a held-out 115-document test split, reached over six hardening rounds - each one required to show zero regressions against the previous scorecard before it could merge.
- Combined a trained detector for the mandatory marks with schema-driven zone OCR for the form fields. The hybrid recovered fields that pattern-matching alone had never handled at all, without losing accuracy on any field that already worked.
- Built the evaluation harness that produces those scorecards, plus a synthetic document generator - so every number published traces back to a committed file.
Stack
PythonPyTorchOpenCVTesseractFastAPIPydanticStreamlitpytest
Read the case study
Client work, code is private. The architecture and the evaluation method are covered in the case study, and I'm happy to walk through the code in an interview.
