Did the SpaceX loop for a senior SWE role in their avionics software group. The system design round stood out compared to other places I've interviewed. Sharing specifics because the usual "design Twitter" advice doesn't fully apply here.
The setup: 60 minutes, one senior engineer plus one principal. I was given a prompt at the start of the round. Mine was something in the space of: design a telemetry ingestion and alerting system for a fleet of devices that produce high-frequency sensor data. Deliberately not memorizing exact wording but the domain was real to their work.
What they emphasized:
Reliability and failure modes first. Before I even finished the component diagram they asked how it behaves when the message queue falls behind. Then what happens if an alert fires twice. The focus was less "can you name the right services" and more "do you actually know how distributed systems fail in production."
Trade-offs, not just the right answer. I proposed Kafka for the ingest layer. They pushed back: why Kafka and not Kinesis or a simpler polling model? I walked through latency requirements, operational complexity, replay semantics. They seemed to like that I had reasons rather than just pattern-matching to the buzzword.
Depth over breadth. I went maybe 60% as wide as I normally would and 2x as deep on the pieces I picked. That felt right in retrospect. They asked detailed questions about the storage layer and compaction behavior, not whether I mentioned a CDN.
At the senior level I'd estimate this is the deciding round. The coding rounds are more pass/fail. System design is where you show if you're at L4 or L5 and above.
One other thing: they did not ask any ML design questions. This was a pure backend/infra loop. YMMV by org.