Just finished my full loop for a senior SWE role (they call it roughly L5 equivalent) and wanted to share specifics on the system design round since I couldn't find much concrete info before going in.
The design session is 60 minutes with two interviewers. Format was pretty open: I got a whiteboard prompt (virtual whiteboard, we were remote) and was asked to design a system that processes and routes large volumes of structured AI-generated outputs. Think something adjacent to their actual infrastructure. Not a generic "design Twitter" type question.
What they actually cared about: Reliability and fault tolerance. One interviewer kept probing on what happens when a downstream consumer goes down. I had to sketch out retry logic, dead-letter queues, and backpressure handling in real time. Latency vs. throughput tradeoffs. They pushed me on whether I was optimizing for p50 or p99 and why. Data consistency. Not distributed systems trivia, but genuinely: if two replicas diverge, what does the user experience look like and how do you surface that?
They did NOT ask me to implement anything in this round. No coding. Just talking through architecture, drawing boxes, defending tradeoffs.
Calibration note: I have about 8 years backend, heavy on distributed systems. The questions felt calibrated for someone who's actually operated systems at scale, not someone who memorized system design books. If you've only done leetcode + book prep, this round will surface that fast.
Duration was 55 minutes with about 5 for Q&A. One interviewer was clearly more senior and did most of the probing. The other took notes.
For prep: I'd focus on real tradeoffs you've made at work rather than trying to pattern-match to textbook answers. They seemed genuinely curious about how I think, not whether I landed on a specific architecture.
Happy to answer follow-up questions on specific areas.