The question was about downloading a model.
It went so-so, even though the question itself was actually the simplest version of it. The main issue was that download and upload shared a 10Gbps bandwidth cap — not sure if the interviewer just picked that number randomly.
The main problem was that early on I spent a lot of time performing — working through and explaining the optimal algorithm. I'd actually already prepared the p2p solution, but I wanted to perform the slow evolution from a basic solution up to p2p. It turned out a few of the suboptimal algorithms in the middle had small issues, the interviewer didn't seem to be following along, and they gradually lost interest.
I got the sense the interviewer's point was that the specific algorithm and topology didn't really matter — you just needed to argue what the minimal bound was. So I gave up on that approach and went straight into designing p2p. Time was a bit short after that; I tried to cover all the deep dives, but I felt the interviewer wasn't very engaged anymore. I didn't really let the interviewer bring up deep dives — I mostly brought them up myself, proactively.
Update (2026-08-05): The next day I got notified that I passed. The onsite has a coding round and a system design round.
Discussion
Loading comments…