Launching today
Paguely
Your blind date with a book. Mood-matched. Mystery first.
22 followers
Your blind date with a book. Mood-matched. Mystery first.
22 followers
Paguely flips book discovery on its head. Instead of scrolling covers and blurbs, you answer a few mood questions and get a cryptic, atmospheric clue about your next read, no title, no cover. You commit blind, then it reveals. Beyond the reveal: a real reading life built in. Spine-view shelf, streak tracking, a mood-based reading DNA profile, and a community feed for sharing reveals. Built with Flutter, Supabase, and OpenAI's GPT-4o-mini.
Such an interesting concept. I tried it and it works like a charm. Getting a book in my hands that I probably would never even take from a bookstore shelf. I will keep exploring more book as soon as I finish this one (Loving Jack btw.). I have just one question.
What stops AI from giving me the same book twice?
@kurtaΒ Thank you for the feedback! Happy you are enjoying it and already found a use for it!
Great question, it's the thing we spent the most time getting right.
Short answer: the engine keeps a memory of every book you've already met, and it treats different kinds of "met" differently.
If you've passed on a book, that's permanent. You said no, and there's no expiry on no. It will never come back to you.
If you've already revealed a book, or saved one recently, it gets pushed down rather than removed. A book you loved is still a legitimate match one day, just never the first thing we hand you.
On top of that, every reveal starts from your mood answers right now, not just your long-term taste profile. Those two are scored separately and weighted, so the same person on a restless Tuesday and a cosy Sunday genuinely lands in different corners of the catalogue. Add a novelty signal that actively rewards books you haven't crossed paths with, and repeats stop being the failure mode.
One thing worth clearing up, since people assume otherwise: for the Reveal we don't ask an AI "what book should this person read." The model helps us aim the search and writes the clues, but the actual pick is deterministic scoring over our own catalogue. That's deliberate. A model asked to name a book will happily name the same famous novel to everybody, and occasionally invent one. Scoring rows we actually hold can't do either.