My personal, preferred way to interview data scientists (besides administering a work sample test) is to talk to them about the most challenging effort they've

My personal, preferred way to interview data scientists (besides administering a work sample test) is to talk to them about the most challenging effort they've been a part of. Start with understanding the overall business context and the technical infrastructure they were working with, and gradually follow through until you understand the key ideas that underpin the innovation. The key is to get them to explain things until you understand the essence at a foundational level - hopefully you understand what combination of the notions of statistical independence / power / dimensionality / algorithmic complexity has birthed novel system or algorithm they built. The process selects for people who have a) already built something substantial, and b) have the communication skills to explain it.