Sharing a failure story from September. The recruiter reached out to me on LinkedIn, and then we scheduled a phone screen. Since this was hiring for a newly formed applied GenAI org, the format was pretty new and I haven't seen it in any other interview write-ups: 30 min of ML fundamentals + 30 min of practical coding.
30 min ML fundamentals, three questions:
- What is a KV cache?
- What is attention, and how exactly are Q, K and V computed?
- What is the difference between L1 and L2 normalization?
30 min practical coding: the interviewer pasted a PR for a feature pipeline and had me review it on the spot. The test cases all passed, but I was asked to analyze what problems it would cause if it landed in production. I caught some edge cases, but I felt like I didn't really answer what the interviewer was after (shrug). At the end we discussed how to serve these features under different real-time requirements.
Two weeks after the interview I got a rejection email, with basically no feedback. XD
Discussion
Loading comments…