Thanks for sharing! Yes, getting good performance out of pgvector with even a trivial amount of data requires a bit of understanding of how Postgres works.