You are building an ML system for a financial platform that must score incoming transactions in real time and help decide whether to approve, review, or block them. The system must learn from historical activity and adapt as fraud patterns and user behavior change.
How would you approach designing a system to handle millions of transactions per second?
Real-time model serving at very high throughputOnline and offline feature store designRisk scoring and policy thresholdingDrift, skew, and production monitoring