Did a senior SWE onsite at Costco last quarter and the system design round was the most interesting part, so worth its own post.
The prompt I got was something like: design an inventory tracking system that syncs between warehouse locations and the e-commerce website in near-real time. Very on-brand for them. You could tell it was a problem their actual engineers wrestle with.
What they wanted to see:
They weren't looking for you to name-drop Kafka immediately. The first thing the interviewer asked was "what are the consistency requirements here" which set the tone. They care about tradeoffs, not tech trivia. I spent probably 15 minutes just scoping requirements and edge cases before drawing anything.
Things that landed well: Talking through the difference between warehouse inventory counts (batch updates, 15-minute lag acceptable) vs. online checkout (needs to be accurate enough to avoid overselling a $3,000 refrigerator) Mentioning eventual consistency and compensating transactions Keeping the design boring and reliable: a message queue, an inventory service with a write-ahead log, a caching layer for read-heavy product pages
Things that didn't matter: Kubernetes specifics Exact AWS service names (they're on AWS but not exclusively) Fancy ML or streaming analytics
Format: 55 minutes, one interviewer plus someone "taking notes" (probably a shadow). Whiteboard on Zoom using their own canvas tool, not standard CoderPad.
Level-wise this was clearly targeting senior/staff. The follow-up questions went deep on failure modes: what happens if the message queue falls behind during a Black Friday spike, how do you recover stale cache entries, etc. Know your failure modes.
If you're coming from a startup background the scale numbers are real but not extreme: they process millions of orders, not billions. Reasonable design is enough. You don't need Google-scale solutions.