Real-Time Fraud Detection at Scale
summary
Architecture uses Apache Kafka for ingest and Apache Flink for stream processing, deploying models horizontally and a custom graph DB plus GNNs for relationship detection. The pipeline claims sub‑millisecond stream latency and 10 ms end‑to‑end prediction, scales near linearly (doubling servers adds ~5% overhead), and reduces memory by ~40% with 3x faster graph queries. Hybrid detection combines ensembles, behavioral signals and NLP, catches coordinated campaigns with as few as 8 accounts, and reports a 78% drop in fake reviews in production.
AI-written summary · about
chapters (10)
- 0:00Introduction to Real-Time Fraud Detection
- 0:56The Challenge of Fake Reviews
- 2:31Building a Scalable Detection System
- 3:58Hybrid Detection Approach
- 9:22Performance Metrics and Effectiveness
- 11:27Coordinated Fraud Campaigns
- 13:22Privacy and Data Protection
- 15:14Scalability and Real-World Deployment
- 18:01System Flexibility and Future-Proofing
- 21:56Key Takeaways and Conclusion
related talks
51mReal-time Fraud Detection: Challenges and Solutions
10mReal-Time Fraud Prevention at Scale
13mBuilding Financial Fortresses: Real-Time Fraud Detection
29mBuilding Resilient AI Platforms: Financial Fraud Detection
47mReal-time Fraud Detection Challenges and Solutions
13mBuild Real-Time Analytics Platforms for Financial Crime Detection
46mData Analytics-Based Detection At Scale
29mReal Time Fraud Detection with Stateful Functions