For you

Ask, and the librarian will pull five books from the stacks to fit your taste — weighted toward the authors you have loved.

The librarian ranks real books already in our library by the authors you have loved (weight 0.5), embedding similarity from a local qwen model (0.3), and genre overlap (0.2), then a local qwen writes the reason (with a deterministic fallback if needed). A book is only ever suggested if it truly exists in the library with a real link, so nothing is invented. Group consensus (on a circle's page) instead picks the one book that maximises the whole circle's taste, blending the members' mean and minimum scores so no one is handed a book they would hate.