Your question is Real-Time Fraud Pattern Detection. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
How would you design a system to detect fraudulent patterns in real-time?
Explain how you would process streaming events, generate risk signals, serve predictions with low latency, and handle delayed or incomplete labels. Address architecture, model selection, evaluation, monitoring, drift, training-serving skew, and failure recovery.