An organization is building a recommendation system that requires low-latency vector similarity search. They need to store and query millions of embeddings. Which THREE technologies are appropriate for this task?