just finished my Zoom L5 loop in May 2026 and the system design round was the most interesting one. sharing notes because the format is a bit different from what I expected coming from a big-tech background.
the round is 60 minutes. first 5-10 minutes are clarifying the problem, then you're expected to own the whiteboard (virtual, via Zoom's own screen share) for the bulk of it. the interviewer I had was senior staff level and clearly knew their stuff. they let me drive but pushed back firmly when I made assumptions without justification.
what they asked: I got a video conferencing reliability / signaling system. very on-brand for Zoom obviously. you need to think about WebRTC, TURN/STUN servers, SFU vs MCU architectures. if you're targeting L5 and you don't at least mention the trade-offs between media relay approaches, you'll get dinged.
for data modeling: they wanted me to think about participant state, session persistence on reconnect, and how you'd store recordings at scale. S3-backed blob store is obvious, but they pushed on the metadata layer and how you'd handle time-based queries efficiently.
what they care about at L5: they're not testing if you know every AWS service by name. they want to see structured thinking, explicit trade-off reasoning, and awareness of failure modes. I said 'what if the signaling server goes down mid-call' and then talked through failover and the interviewer visibly relaxed.
no deep dive into leetcode-style data structures in this round. it's purely design.
tips: know your distributed systems basics. CAP theorem will come up. think about horizontal scaling before they ask. and given the product, know at least a little about real-time communication protocols. you don't need to be an RTC expert, but 'I'd just use websockets' for a 1000-person call will not land.