Back to all selected work
Financial Systems · Applied ML·2024·built
FRIP — Real-Time Fraud Scoring
Streaming transaction fraud scoring on Kafka with ML anomaly detection.
JavaSpring BootApache KafkaRedisMachine Learning
Overview & Context
High-throughput risk scoring service processing financial transaction streams.
The Problem
Evaluating machine learning models synchronously on financial transaction paths introduces unacceptable latency without intelligent caching.
The Architectural Approach
Connected Kafka event streams to a Spring Boot scoring service with multi-tier Redis caching to bypass duplicate inference on recurring merchant profiles.
Key Technical Highlights
- Kafka transaction event streaming to Spring Boot scoring microservice
- Anomaly detection scoring with sub-10ms evaluation target
- Redis cache layer preventing redundant model inference
Next Case Study
View projectPingPanda
Send product events to Discord in real time.