About

Visual Style Twin is a production-grade, image-based fashion recommendation system built entirely on Zerve.

It enables “shop by visual style” by learning clothing aesthetics directly from images and delivering real-time, diversity-aware recommendations through a deployed API (since don’t have access to API deployment) or scheduled batch.

This project demonstrates a complete analytical system: data ingestion, visual embedding learning, rigorous validation, baseline comparison, production deployment, and scheduled refresh workflows.

BUSINESS PROBLEM

E-commerce platforms struggle with visual discovery. Keyword filters and categorical navigation fail to capture subtle visual cues such as color composition, texture, silhouette, and styling. As a result:

• Users abandon searches early
• Cross-category discovery is weak
• Visually similar alternatives remain hidden

Visual Style Twin addresses this by recommending products purely on visual similarity, enabling intuitive exploration and improved engagement.