RowRecon
Convert a scanned bank statement to CSV locally
When a PDF has no usable text layer, browser-local Tesseract OCR may produce REVIEW rows.
Local OCR is not universal. Clean scans can be usable. Poor scans should fail or stay in review rather than guess HIGH amounts.
The generic local parser may attempt the file. Bank-specific support is not claimed.
RowRecon shows users where extraction may be wrong before they rely on the export.
Convert locally
Upload a PDF, review flagged rows and reconciliation, then export CSV or Excel. The file is not sent to take payment.
Upload a statement PDF
PDF only. Maximum 15 MB and 40 pages. Scanned pages use local OCR in this browser (up to 20 pages) when the file has no usable text layer. Review uncertain rows before export.
Try a sample
Synthetic fixtures only. Samples run through the same local extractor. Output is not hardcoded.
Report a problem
Sends privacy-safe metadata only (parser, counts, recon status, replay fingerprint). The PDF never leaves this browser. Download a redacted layout replay to recreate the issue in tests.
How it works
- Choose a statement PDF (or a synthetic sample).
- Extraction runs in this browser, including local OCR when there is no text layer.
- Uncertain rows stay visible. Click a row to see the source region.
- Where totals exist, reconciliation checks whether balances tie out.
- Export CSV or Excel after review.
FAQ
- Is OCR remote?
- No. Tesseract.js runs in a Web Worker in your browser. Trained data is served from this origin.
- Why are OCR rows not HIGH?
- OCR-derived rows are capped at REVIEW so uncertain scans are not promoted just to raise a headline accuracy number.
- What are the OCR limits?
- Up to 20 pages, 8 million pixels per page, 2200px edge, with timeout and cancel. Larger or encrypted files fail closed.