How would you design a scalable vector similarity search pipeline that handles 10M+ high-dimensional embedd…

How would you design a scalable vector similarity search pipeline that handles 10M+ high-dimensional embeddings per day while guaranteeing sub-second latency for top-k retrieval?