had the system design round at RenTech last fall for a senior SWE role. sharing notes because the format is a bit different from what you'd get at a pure product company.
the setup: 45 minutes, one interviewer (senior engineer), no whiteboard gimmicks. they gave me a prompt upfront and let me drive. mine was something like: design a low-latency data pipeline that ingests market data and routes it to multiple downstream consumers. very trading-adjacent. not a generic "design Twitter" round.
what mattered: latency over everything. when i said "we can use Kafka here" they immediately pushed: what's the tail latency, what happens when a consumer lags, how do you isolate backpressure. at a product company they might nod at Kafka and move on. here they assumed i actually knew how it worked. failure modes in detail. they wanted failure scenarios, recovery strategies, what happens under network partition, how the system degrades gracefully. the CAP theorem came up naturally, not as a gotcha. operational reality. monitoring, alerting, how you'd debug a latency spike at 4am. clearly they have an on-call culture and they want to see you think operationally, not just architecturally.
what didn't matter much: rough estimates and back-of-envelope math around scale. they seemed more interested in whether you understand the tradeoffs of your choices than whether you can estimate QPS correctly.
how i'd prep: honestly, reading through distributed systems literature (the DDIA book covers most of the conceptual ground) and knowing your streaming tech stack cold. if you've run Kafka or Flink or similar in prod, talk through the failure cases you actually hit.
the interviewer was genuinely engaged and knew the space deeply. felt more like a technical discussion between peers than an evaluation. just one where the peer is probably smarter than you.