finished my capital one senior SWE loop last month. three onsite rounds plus a take-home. the system design round was the one i stressed about most, so here's what actually happened.
the prompt was something like: design a fraud detection pipeline for real-time card transactions. not a unique prompt but extremely on-brand for a company that processes hundreds of millions of transactions. the interviewer was an E6 equivalent who had clearly run this prompt a hundred times.
what they cared about most: how you handled scale. they wanted specific numbers. don't say "lots of transactions", say "5,000 TPS at peak with burst capacity for holiday weekend" data consistency trade-offs. you're in fintech so eventual consistency needs justification. they pushed back hard when i said kafka with at-least-once delivery without explaining how i'd handle duplicate fraud alerts latency requirements. the fraud check has to happen before the transaction approves, so sub-100ms at p99 matters. they wanted me to talk through the call stack
what they were less focused on than i expected: deep leetcode-style optimization of anything the exact infrastructure choices (AWS vs GCP, kafka vs kinesis) as long as you could reason about the trade-offs caching layers (i brought it up, they didn't push me on it)
the 45-minute format: 5 min intro/clarifying questions, 30 min design, 10 min "where would this break" scenario. that last part is where people struggle. have a failure story ready.
prep that actually helped: distributed systems primer, designing data-intensive applications (the kleppmann book), and one mock run with a fintech-familiar friend. the domain matters here. if you have fintech background, lean into it.
one thing: they DID ask about my comp expectations at the end of the design round. weird timing but heads up.