Tool docs that helped me: vector DB benchmarking↗ arxiv.org
41
One more thought: validate assumptions with a small A/B test if possible.
Watch out for edge cases: duplicates, missing fields, and race conditions.
Small tip: document the decision so future-you remembers why you picked it.
Agree. Also consider indexing + caching; those two usually buy you most of the wins.