First round was a take-home Leetcode Medium OA. Second and final round were two virtual back-to-back 45-minute technical interviews: you talk about your past experience, they ask about metrics/eval/general ML knowledge in the context of these projects, and they talk about what they're team works on. At the end there's a pretty short LeetCode question. Less about syntax, and more about approach and being able to articulate directions/tradeoffs more than being able to implement. No debugging