MyBookRecommendation
Describe a mood, a trope or a book you loved and get real books picked for you, each with a reason.
Visit mybookrecommendation.riyaaz.workers.dev- Type
- AI product
- Role
- Designed the whole system end to end
- Built with
- Gemini API, Cloudflare Workers, Open Library API
- Result
- Every pick verified
The problem
Bestseller lists tell you what everyone is reading. They cannot tell you what you will love next.
Asking an AI model directly has its own catch: models confidently recommend books that do not exist. For a recommendation product, one invented title destroys trust.
What I built
A reading matcher that takes plain words, like a mood, a trope, a page count or a book you loved, and returns picks ranked by how well they fit, each with a one-line reason, its length and where to buy or borrow it.
Every suggestion from Gemini is checked against a real library catalogue, Open Library, before it reaches the screen. If a book cannot be verified, it is never shown. The model does the understanding; the catalogue keeps it honest.
Running it as a real product
Around the matcher: "Books like…" search, recent Booker, Pulitzer and National Book Award winners, hand-picked lists for very specific moods, saved books, and a Sunday email with five books for the weekend.
It runs on Cloudflare Workers and earns through Amazon and Bookshop.org affiliate links, disclosed on the site, with no sponsored placements.
What it taught me
The hard part of an AI product is not calling the model. It is making the output dependable enough that the interface can trust it, every time, for every kind of request.