Whenever you're ready, walk me through how you'd approach this.
Explain how you would design a system under strict cost and latency constraints for a client scenario.
Treat this as a practical architecture exercise. Clarify the client’s product objective, traffic and data characteristics, latency and availability targets, budget, and compliance needs before proposing a design. Address the end-to-end ML lifecycle, including batch and online paths, model serving, evaluation, monitoring, rollout, and failure recovery. Explain how your choices would change if the client prioritized lower cost, lower latency, higher accuracy, or faster feature freshness.