I broke the problem down by clarifying requirements first, then proposed a high-level architecture using load balancers, stateless services, and partitioned data stores. I discussed consistency tradeoffs, caching strategies, failure handling, and how the system could scale horizontally. I also walked through potential bottlenecks and how I would monitor and mitigate them.