Whenever you're ready, walk me through how you'd approach this.
Design a key-value store.
Asked in the onsite stage. Clarify the API and consistency expectations, then explain how you would partition, replicate, serve, monitor, and recover the store. Address online reads and writes, persistence, hot keys, node failures, rebalancing, and capacity planning. If relevant, discuss how the design could support low-latency ML feature lookups without introducing training-serving skew.