Whenever you're ready, walk me through how you'd approach this.
How would you use vector databases in a retrieval system?
Treat this as a practical pipeline design exercise. Explain ingestion, chunking, embedding generation, indexing, metadata storage, incremental updates, deletes, retrieval, backfills, data quality, monitoring, and failure recovery using Google Cloud services where appropriate.