Ask HN: How to run technical interview for deep learning engineers

The startup I work for is growing rapidly, and I am in charge of building a team to develop models to analyze medical imaging data. This is my first time in charge of the entire interview process. How should I structure my technical interview to assess if they are capable while not making the applicant too stressed or uncomfortable?

What I do currently is pose a situation similar to the work we do, and let them take the lead and describe how they plan to approach the problem. I'll then bring up problems they could encounter and ask how they'd address them. I don't like to ask any questions that you'd see in articles about preparing for a machine learning interview, since I feel it only selects for people who cram for the interview and tells me little about their ability to just get stuff done using any resources available.

How would you improve my methodology?