Just ask them where they needed to solve a problem with an algorithm, why/what they did, how they implemented, tested, validated. What other problems could use the same approach, how could the problem be changed to invalidate assumptions, and so on. etc.

(While writing I realised that this fails in cases where the interviewers are not competent enough to follow along in random directions, which then gives the rationale for a leetcode quiz - basically a battleship-game that the interviewer prepares in advance so that he can pace the discussion and grade the candidate.)