Your question is Two Tower vs TikTok Recommender. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
Can you compare the advantages and/or disadvantages of a Two Tower recommender to something more complex like TikTok's recommender?
Asked in the tech screen stage. Discuss the comparison as an end-to-end ML system design problem, including retrieval, ranking, serving, evaluation, and operational risks.
Address model expressiveness, candidate generation, feature freshness, inference cost, cold start, training-serving skew, and when a two-tower architecture is preferable to a more complex multi-stage system.