Whenever you're ready, walk me through how you'd approach this.
Design a data pipeline or serving architecture for a specific ML application.
Choose a realistic ML application and explain the end-to-end design. Cover requirements, scale assumptions, data ingestion and labeling, training and feature computation, online or batch inference, evaluation, monitoring, and failure recovery. Make tradeoffs explicit, including latency, cost, freshness, training-serving skew, and feature drift.