Author here. Yes, it actually does work. If I’m ever indexing fewer than ~100k documents/paragraphs I would use something like this approach rather than reaching for a vector database.