Your question is Vector DB vs Custom Indexing. 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).
What are the tradeoffs between using off-the-shelf vector databases versus custom-built embeddings and search indices?
Compare the approaches for a production RAG system, including retrieval quality, latency, infrastructure cost, scalability, metadata filtering, operations, and security. Explain how you would evaluate the decision offline and online, and when a hybrid design would be appropriate.
Your answer should distinguish the embedding model from the index and database layers, identify major failure modes, and make assumptions explicit.