This project involved designing a real-time fraud detection architecture for banking transactions within a big data environment. The solution was tailored to handle the massive transaction volume of one of Brazil's largest banks while ensuring scalability, reliability, and speed.

Key aspects of the architecture include:

High-Throughput Data Ingestion: Designed to ingest a continuous stream of high-scale transactions in real-time, leveraging streaming technologies to handle the bank’s transactional flow.

Data Enrichment in Streaming Windows: Implemented a system to enrich transactional data with auxiliary information, such as customer history or external risk signals, within defined processing windows to maintain real-time performance.

Queue Integration for Consumption: Fraud scores were written to message queues, enabling seamless consumption by downstream applications, including alert systems and response workflows.

Historical Storage for Analysis and Audits: Ensured all processed data was stored in a structured format for use in analytics, manual reviews by fraud analysts, and future audit requirements.

This architecture empowered the client to detect potential fraud with low latency, provided tools for analyst investigations, and ensured regulatory compliance with robust data traceability.