Build the like button for a social product at scale: every post, video, story, and tweet carries a heart. A user taps it once to like, taps again to un-like; everyone else sees a count that updates in near-real-time. The system is read-dominated by 100–500×, hot-keyed on viral posts, must give the liker read-your-write on their own heart, and accepts approximate counts everywhere else.

The interview answer ("one Redis INCR, one Postgres row, done") survives ~10K writes/s and 100K reads/s on a single non-viral product. We're sizing for 2B likes/day with single-key viral peaks at 50K writes/s and read amplification 100–500×, and we have to defend it at 3am when Ronaldo posts.