Whenever you're ready, walk me through how you'd approach this.
Design a high-level system capable of handling millions of events from thousands of sources.
Clarify the event contract, delivery guarantees, processing requirements, and whether events require real-time ML inference or batch analysis. Explain the architecture, scaling assumptions, storage strategy, observability, and failure handling. Discuss how you would evaluate and operate any ML component, including feature drift and training-serving skew.