System Overview
This document describes the end-to-end flow for processing bank statements across multiple banks. Each bank has its own statement structure (page layout and transaction placement). The system converts statements into HTML templates, applies bank-specific rules, audits incorrect statements using AI, fixes detected issues, and generates corrected statements.
🎯 Key Objectives
Multi-Bank Support
Handle 12+ different bank statement formats
AI-Powered Audit
Automated validation and error detection
Template Generation
Standardized HTML output formats
Error Correction
Automatic fixing of detected issues
High-Level System Flow
1
Template Preparation
- Each bank statement format is converted into a static HTML template and saved in backend
- Templates are divided into pages (Main, Transactions, End Statement) based on bank-specific layouts
- A rule set is defined for each bank (page order, transaction start/end, validations)
2
User Input
User uploads a PDF bank statement (potentially incorrect or inconsistent).
3
Data Extraction
The system extracts structured data from the PDF, including:
- Account details
- Statement period
- Balances
- Transactions
4
Rule Evaluation (AI-Powered)
- AI identifies the bank type
- AI validates extracted data against bank-specific rules, such as:
- Transaction placement
- Page boundaries
- Missing or inconsistent fields
5
Audit Report Generation
AI generates an Audit Report containing:
- Detected issues
- Rule violations
- Formatting or data inconsistencies
6
Statement Fixing
Based on the audit report:
- Incorrect data is corrected
- Layout issues are fixed
- Data is re-mapped into the correct HTML template
7
Final Output
A corrected bank statement is generated from the HTML template.
Output can be exported as:
- PDF
- HTML (for review or further processing)
⚠️ IMPORTANT NOTE
Model Limitation: Maximum processing capacity is limited to 10 Pages per statement.