Position: Applied Science: PhD Internship Opportunities - Redmond
HR emailed to invite me to interview, no OA.
VO1 (02/04): a 30 minute simple chat, talked about past projects, asked about some details.
Then scheduled three back-to-back VOs on the same day (02/19), each 45 minutes, with a 15 minute break between adjacent interviews.
VO2: an applied scientist, also asked about technical details of past projects, this time in more depth, plus some ablation study design questions.
VO3: a senior architect, mainly asked about current NLP techniques like RL and agents, then mixed in some ML questions, like the role of precision/recall, ROC/AUC, the difference between L1/L2 regularization and when to use which, and classification metrics. With maybe 10 minutes left, he threw in a coding question, and said it was fine if I didn't finish in time. It was easy difficulty though — roughly: given a list like [-1, -2, -4, -5, 7, 10], find the minimum difference between numbers, and output the corresponding pairs: [[-5, -4], [-2, -1]].
VO4: a software engineer, tested one coding question, but not an algorithm one: suppose you have 10 files, each file contains a picture of an animal, and you can call an API that returns k files — write pseudocode to compute precision and recall for dog pictures among the returned results. The follow-up was: what if the API's return is abnormal (e.g. it returns None, or doesn't return exactly k files) — how do you handle that and compute the metrics? Then they asked how I'd design a RAG system. Since I don't work on RAG, I answered this one so-so, tying it in with agentic tools.
Got the official offer email on 02/23.
This was also the first time I went through this kind of three-back-to-back interview format, and I was pretty wiped out afterward. Overall I felt every interviewer was quite engaged, all tech guys, gave good feedback, and if there were details I didn't get across clearly they'd let you explain further. This might also depend on the team — for example I didn't get a behavioral question round, and interviews can apparently differ a lot between teams.
Discussion
Loading comments…