Rethinking book discovery through recommendation systems and UX

Year

2026

Industry

Books, Startup, ML

Client

Unread

My role

Creator

A work-in-progress platform combining curated discovery, interaction design, and a proprietary ML recommendation system.

UX/UI

AI Coding

Web App

Database

Currated Feed

Business Strategy

How it works?

When readers rate books, a unique personal profile is created. Every time they ask for a recommendation, Unread analyze at least 20 different "style levers" to choose and rank the top 10 books. It's been benchmarked as 20 times better at predicting what people would enjoy reading next than Gemini 3.1.

For the Data Scientists: Unread proprietary two-stage model uses a Two-Towers retriever paired with a FiBiNET ranker to ensure high-precision matching.

More details soon!

In the meantime:

Mastodon