The interview process was 4 interviews in total, the first interview was with the hiring manager and was a very general review of resume and experience. The next 3 interviews were with various other technical members of the staff, and focused more on technical assessment. Oddly enough, in all four interviews, none of the questions really targeted intuition around deep learning for computer vision, and this position was predominantly for computer vision. The first technical interview was pretty awkward. The interviewer opened a google doc and a programming problem, then told me to code directly in the document. I asked him if I could at least put the code in an editor (since nobody ever codes in google docs) but the interviewer said 'no'. So I wrote the program in google docs, fighting formatting the whole time, and was not allowed to test my code. Most companies use leetcode which provides a decent editing environment and also allows you to run your code, even during interviews. But that was not the case here. So they essentially put me into a scenario that that would never occur in the real world, asked me to solve a contrived problem, then disallowed me from checking my solution and presenting it confidently. I'm not sure putting candidates in an irregular circumstance and asking them to solve a problem without typically available resources is an effective way to gauge their abilities. The next interview was with the software architect. He also asked some technical questions about deep learning, but they were only surface-level questions about hyper parameters and optimizers. We didn't talk at all about deeper topics like network architecture or what a network learns. At one point in this interview, and perhaps most surprising, he put a conspicuously labeled python function in front of me and asked me what it did. It was a recursive function that computed the FFT of a series, not something very commonly known or taught in deep learning for computer vision. So it was weird that he would ask about something so irrelevant. I think we was just trying to assess me in terms of what his background was, which happened to be somewhat irrelevant to the position as advertised. The final technical interview was with the hiring manager who does deep learning. I was expecting this interview to go better than the previous two. The interviewer ended up asking some really obscure questions about applying deep learning models to challenging circumstances. It was clear that the interview questions were derived from real challenges he was facing in his job, which was fine. However, the difference between the interviewer (him) and the interviewee (me) in this circumstance is that he's known about these problems for some time, and has been able to research, experiment, and analyze these obscure problems. I, as the interviewee, am being put on the spot to both understand these obscure applications and also describe details about why the problems occur and how to solve them. Again, no real focus on the fundamentals of deep learning for computer vision, just very specific questions about obscure problems. I'm not sure this process is going to highlight good problem solvers with solid intuition more than it is going to find specific candidates who may have dealt with hyper specific cases. But I wish them good luck.