This follows my earlier post about the recruiter round. Here are the two rounds of stage one. Background: about 4 years of SWE experience. Manager interview on 11/13, coding interview on 11/16. There are still a design round and a final round after this, and I'm waiting to hear back.
Manager interview (11/13, 1 hour)
The interviewer was the hiring manager of the team. He came up as an engineer and spent years at Amazon before joining Docker earlier this year.
About the team: fully remote, no requirement to go to an office (there are offices in Seattle, the Bay Area and New York if you want to go). Right now there are 2 engineers, 3 more about to join, and the goal is to grow to 8. They have a standup every morning, which personally felt a bit micromanage-y to me.
The questions were mostly behavioral and deep dives into my experience, and he asked a lot of them. The only one that stumped me: "If a PM proposes a requirement and you clearly disagree, how do you handle it? Do you have a story like that?" My current team has no PM at all, so I froze. I told him so and tried to tell a similar story, but he said no need and just skipped it. A reminder to everyone: even if your current company has no PM, prepare a story about disagreeing with a PM or stakeholder ahead of time.
Coding interview (11/16, 1 hour)
Both interviewers were from other teams, one principal and one staff, not people from the hiring team.
Format: no shared IDE. I shared my screen and wrote the code in my own VS Code. The interviewers said explicitly that I could use AI (to write tests, to fill in code, either is fine).
The original problem:
You're building an API rate limiter to prevent clients from sending too many requests in a short period of time.
Each client is identified by an id (e.g., an IP address or API key).
Your task is to implement a function that enforces a rate limit per client ID.
How it went: first we talked through the common algorithms and their tradeoffs out loud (fixed window, sliding window log, sliding window counter, token bucket). I picked sliding window log to implement first, because it is accurate and the logic is straightforward. The downside is that you store timestamps, so storage is the bottleneck. Talking and writing took about 30 minutes in total. Then I had the AI generate test cases (I described out loud what to test: invalid input, edge cases, the rate-limiting logic), and they ran fine.
Follow-up: what do you do about storage with 1M clients? That led to the sliding window counter, and I implemented that too. I got stuck halfway, right at the point of comparing the pre-estimate against max_request, where I suddenly went blank. The interviewer said I could use AI, so the AI filled in the part I was stuck on and I finished without trouble. I was supposed to have 10 minutes at the end for my own questions, but only 2 minutes were left, so I asked one question and that was it.
Overall experience
Overall it felt really comfortable. The interviewers were very open about AI, and it felt like they cared more about the thinking and how you explain tradeoffs. The only hurdle: I really could not understand how one of the interviewers expressed himself. I asked a few times in the middle and still didn't get it, so I just kept writing code. The other interviewer communicated very smoothly.
No result yet.
Discussion
Loading comments…