Interviewing for data science is incredibly hard - first of all it’s a super broad field, incorporating everything from making models, to cleaning data, to doing cutting edge research on any one of the many current domains - Transformers, alternative architectures, more classical machine learning like random forests, model compression, etc.<p>One thing I really do believe is that to be an actually effective research team member, that does not require a lot of outside help, is coding skill. If you can’t coherently program your idea, in a way that others can use in a bug free way, it means someone else has to troubleshoot and fix the spaghetti.<p>I like to split my interviews into two parts - one decently tricky programming problem, and one “design” type question. Except instead of system design, it’s research design. We have this problem, how would approach making a hypothesis, testing it, and expanding on it from there.<p>Finding someone who can do both is _hard_ though!