Matcha Bank Reconciliation System
What
CS Capstone for UC'26. Matcha is an automated bank reconciliation system. Aim to reduce manual bank statement and receipts matching while keeping human in the loop for review and reconciliation override.
Why
Manual bank reconciliation is a fragmented, time-consuming and error-prone process.
How
Matcha has 3 core components:
- Documents parsing pipeline using Visual Language Model (VLM)
- Reconciliation matching algorithm
- Centralized dashboard for viewing and exporting reports
Notable features:
- VLM parsing pipeline replace traditional OCR + NLP pipeline with better accuracy.
- AI generated reconciliation summary on unmatched transactions and provide next step.
- Per account book document retention policy for better financial document security and operation cost.
Stacks:
- Vision Language Model (VLM) parsing module via AWS Bedrock
- AWS infastructure for hosting and message queues (EC2, SQS)
- FastAPI backend
- AWS RDS Postgres
- Next.js + Shadcn UI
Results
Matcha shorten the time to process a standard 30 lines statements from 45 minutes to just under 5 minutes while also capable of processing other document in parallel.