Well, I have a brute force strategy for pgvector working reasonably well. Individual, partial indexes. It works for all those queries with category_id=<> clause

Well, I have a brute force strategy for pgvector working reasonably well. Individual, partial indexes. It works for all those queries with category_id=<> clauses. You only need an index for larger categories, for categories with rows below a threshold you dont need index a KNN/dot product would work.