rd.
AI × FashionShipped

Personalized discovery for a visual-first category

Helping shoppers find pieces they'd actually buy in a category where taste is the product. Recommendation, shoppable feeds, and AI-powered styling — but in a way the user trusted instead of resented.

My role
Product + AI scoping. Worked on the recommendation stack and the discovery UX.
The insight
In taste-driven categories, the recommendation engine has to be *visibly considerate* — not just accurate. Users forgive a wrong suggestion if the system shows it understood why they might like it. Black-box "for you" feeds train distrust.
Scale signal
Consumer scale; visual-first.

The AI in this category is less about novel models and more about restraint — knowing what not to recommend, and surfacing reasoning the user can interrogate. Anonymized for now; happy to discuss the pattern privately.

Updates
  1. Sep 2024
    First version of the embedding-driven outfit recommender shipped.
  2. Feb 2025
    Migrated from pure-vector ranking to a re-ranker that weighted "why this was suggested" explanations.