just got through the senior SWE (think L5 equivalent) system design round at NBCUniversal's platform engineering group. writing this while it's fresh because i found basically nothing useful when i searched.
the prompt: design a live video streaming ingestion pipeline that can handle millions of concurrent viewers. they're obviously not doing this from scratch -- the framing is more 'how would you architect this if you owned it'
what they actually cared about: scalability at real media scale. they mentioned peacock viewer spikes during NFL games and asked how i'd handle that specifically. CDN edge caching, adaptive bitrate, origin failover. latency vs. throughput tradeoffs. i spent a good chunk of time on the ingest vs. delivery path distinction and they lit up at that. monitoring and observability. more than i expected. they asked how you'd detect a degraded stream from the platform side before users start tweeting about it. failure modes. what happens when a transcoding node dies mid-stream.
format: 45 minutes with one interviewer, shared whiteboard (virtual). they let me drive but jumped in with constraints and new requirements pretty often. it felt collaborative, not adversarial. i've had system design rounds where the interviewer just stares at you for 45 minutes and this was not that.
what i'd prep: distributed systems fundamentals obviously. but more specifically, look into CDN architecture, HLS/DASH protocols (you don't need to know them cold, but knowing what adaptive bitrate means helps), message queue patterns (kafka was well-received), and basic DB sharding. this isn't a fintech system design -- think availability and throughput over strict consistency.
i'm not a media tech person by background (SRE at a cloud company). knowing the infra patterns transferred fine, i just had to map them to the streaming context.
verdict: one of the better system design rounds i've done this cycle. they were interested in my reasoning, not just the diagram.