Real-time Fraud Detection Challenges and Solutions
summary
Claims a working Hazelcast Open Source pipeline that processes Kafka events in-memory with SQL enrichment and sliding windows to hit millisecond SLAs, not batch latency. Demo runs locally from a jar, reads one Kafka topic once, fans data to multiple Python/Java models, joins streaming and historical maps for enrichment, and outputs decisions; deployment options are host, edge, farm. Targets are sub-100ms (some banks 20ms) and five nines reliability while reducing false positives >$100M per year.
AI-written summary · about
chapters (10)
- 0:00Introduction to stream processing
- 2:30Challenges in production systems
- 6:18Defining real-time requirements
- 10:13Fraud detection architecture
- 13:31Business impact of false positives
- 16:53Stream processing vs batch storage
- 20:52Simplifying the technical stack
- 25:17Designing the system architecture
- 31:28Demonstrating the technical demo
- 37:29Advanced enrichment and joining
related talks
51mReal-time Fraud Detection: Challenges and Solutions
27mBeyond Logs: Real Time Anomaly Detection with Machine Learning
13mBuilding Financial Fortresses: Real-Time Fraud Detection
25mReal-Time Fraud Detection at Scale
27mArchitecting Sub-Second Analytics Pipelines
33mReal-time Trade Data Processing and Machine Learning at Giga-Sca...
13mBuild Real-Time Analytics Platforms for Financial Crime Detection
10mReal-Time Fraud Prevention at Scale