Mpesa Wrapped
Problem
M-Pesa users in Kenya receive transaction statements as PDFs with no way to analyze their spending patterns or get insights into their financial behavior.
Approach
Built a document processing pipeline using AWS Textract to extract transaction data from PDF statements. Designed a classification system that categorizes transactions and assigns users spending personality profiles based on observed patterns.
Implementation
- Spring Boot backend with REST APIs
- AWS S3 for document storage, Textract for OCR
- Redis for caching, RabbitMQ for async processing
- WebSocket connections for real-time progress updates
- MySQL for structured transaction data