System design interview question
Design a Distributed Fraud Detection and Risk Scoring Engine like Stripe's Radar at Scale
Design a real-time fraud detection platform that evaluates payment transactions and user behavior to assign risk scores and make accept/reject decisions within 100ms. The system must process 50K transactions per second globally across multiple payment methods (cards, wallets, ACH), support rule engines with thousands of rules, machine learning model inference, historical pattern matching, and device/IP reputation lookups. Maintain 99.99% availability with sub-100ms latency SLAs, handle geographic distribution across 6+ regions, support multi-tenant isolation (each merchant has custom rules/thresholds), and provide real-time dashboards showing fraud trends. Key constraints: extremely low false positive rates (merchant churn on over-blocking), regulatory compliance (PCI-DSS, GDPR), graceful degradation under load, and ability to rapidly iterate ML models without service downtime.
Practice this question step by step with an AI interviewer that reviews your requirements, your architecture and your trade-offs, and scores the design.