Whenever you're ready, walk me through how you'd approach this.
How would you build and monitor a system for embeddings generation and vector search at scale, including indexing, freshness, and latency trade-offs?
Clarify the workload, retrieval quality goals, freshness requirements, and latency and availability targets before proposing an architecture. Cover offline and online embedding generation, vector index construction and updates, query serving, monitoring, evaluation, capacity planning, and failure recovery. State assumptions explicitly and explain the trade-offs among recall, freshness, cost, and p99 latency.