Capture
Payment-proof images sent on WhatsApp are detected and downloaded automatically.
Document AI
OCR Payment Pipeline
WhatsApp payment proofs are detected, read, and matched to the right account — automatically, in production.
OCR
BCAMandiriBRImoBNIBSICIMB
6 bank formats · 1 pipeline
Overview
Residents send payment-proof screenshots to WhatsApp. This pipeline watches those messages, downloads each image, reads it with an OCR engine, pulls out the amount and the sender, matches it to the right account, and posts the payment. What used to be a person reading bank screenshots one by one is now a background process.
Outcome
6
bank layouts parsed
ONNX
PaddleOCR inference
Live
running in production
How it works
Read
Text is extracted with an ONNX / PaddleOCR engine tuned for Indonesian m-banking layouts.
Bank parsing
Six bank formats are handled: BCA, Mandiri, BRImo, BNI, BSI, and CIMB.
Name resolution
Amounts and sender names are matched to the right account with fuzzy name resolution.
De-duplication
Duplicate proofs are filtered out before any payment is posted.
Self-learning log
A self-learning log keeps improving name matching over time.
Stack
Languages
- Python
- Node.js
Document AI
- ONNX Runtime
- PaddleOCR
Matching
- Fuzzy name resolution
- Self-learning log
Delivery
- WhatsApp Business automation
- Google Sheets
Outcome
No more manual reading
Transfer screenshots are read by the pipeline, not by a person.
Correct account every time
Fuzzy name resolution maps senders to their unit instead of guessing.
Improves with use
The self-learning log makes matching better as corrections accumulate.
Want the same for your operation?
Tell me what runs by hand today. I will tell you whether it can be automated.