Good introductory interview, heavily theoretic traditional ML test for first assignment (website misleading), long wait afterwards. Second stage 90 minutes, interviewer says to expect system design questions, but after heavy preparations, there are none - feels more like NLP exam with questions focused more around transformer and LSTM, with only a glancing view at prior projects / real-life ML cases.
Post-interview feedback seemed pretty condescending, recommending to work on deploying, maintaining, and monitoring the models... problem is, we didn't really discuss this, so it's unclear where the feedback came from.