Whenever you're ready, walk me through how you'd approach this.
Can you walk us through a system design for a specific problem?
Choose a concrete ML product problem, clarify its users and success criteria, then explain an end-to-end design. Cover scale assumptions, data collection, training, retrieval and ranking where relevant, online versus batch serving, evaluation, monitoring, and failure recovery. Treat latency, availability, cost, feature drift, and training-serving skew as first-class constraints.