Replit · Primly Community

Replit senior / L5 system design interview, what to expect

market_realist · 4 replies

Went through the senior SWE loop at Replit in early 2026. Going to be specific because the threads I found before my interviews were pretty thin.

System design was one 60-minute round, one interviewer (a staff eng on their core editor infra team). No co-interviewer shadowing. The prompt was something in the ballpark of: design a collaborative code execution environment that multiple users can share in real time. Obviously no surprise given their product.

What they care about: Latency and concurrency. They drilled pretty hard on how you handle N users typing simultaneously without trampling each other's state. CRDT vs OT came up. I don't think you need to have built one, but you need to know what they are and why you'd pick one. Sandboxing and isolation. Think container-per-session vs a shared pool. Resource limits, escape risks. This felt almost as important as the collaboration piece. Scalability of execution. If you're spinning up code execution per user, what happens when 10,000 people are in a classroom doing the same thing at once? They asked about burst capacity pretty explicitly.

I'd call it medium-hard. Not "design Twitter" where every interviewer has a different rubric. The prompt was focused enough that I could actually go deep. Felt fairer than most system design rounds I've done.

One thing that surprised me: they didn't ask much about storage or DB schema. The focus was squarely on the real-time and compute side. If you're prepping, I'd prioritize WebSockets, OT/CRDT basics, container orchestration concepts over your usual SQL/NoSQL tradeoffs.

Total loop was 4 rounds. System design was round 3. Got an offer at what I'd call a mid-senior L5 equivalent comp, fully remote. Happy to answer specifics if you're preparing.

4 replies

frontend_fran

This is really helpful. Did they expect you to whiteboard (virtual) the CRDT structure in detail, or was naming it + explaining the tradeoff enough?

remote_swe_42

Naming + tradeoff was enough for me. I sketched a rough operational transform diagram but I don't think they were grading the precision of my boxes. The conversation moved pretty fast once they saw I understood the core tension.

infra_ines

Container-per-session at scale is where I'd immediately start worrying about cold start latency. Did that come up? Curious how they think about pre-warmed pools vs on-demand.

jp_newgrad

Thanks for this. Do you know if the system design prompt changes a lot by team or is the collaborative IDE scenario fairly standard?