179 questions
No questions match those filters.
← All questions
Retrieval & EmbeddingsHard· Latency· Scaling· System design
Corpus is about to grow 10×. One engineer wants to wide...
This is one of the questions in the full AI/ML interview bank. Pro unlocks all 1789 questions; Premium includes the same bank plus the highest daily Practice limit.
See plansWidening k on a flat index will cause latency to regress, and building a full three-stage funnel (reclustering and validation) is too time-consuming for a single sprint. The most pragmatic solution is to add a sparse pre-filter to the existing dense stage, which reduces the number of items the dense stage must process, thereby maintaining recall without exceeding the latency budget.