Your question is Vector Search for Massive Datasets. 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).
How would you implement a vector search index for massive datasets with low query latency?
Discuss the design as a practical production system. Cover index construction and updates, approximate nearest-neighbor algorithm selection, sharding and replication, online query serving, capacity planning, evaluation, and failure recovery. Explain how you would handle evolving embeddings, deletes, freshness, memory constraints, and training-serving consistency.