I always imagined there would be a more personal way to shop. Personalization was the single most consistent driver of both higher conversion and a better user experience. I had started working on this concept with Self, a personal memory hub that could share your context with any shopping website you visited. Imagine going to Macy's and, without creating a new account, it already knew the clothes you loved, your height, your skin, and could make better recommendations with that information.
Rylo is the fashion shopping app that gets you. Shop 40,000+ brands and secondhand marketplaces to find anything ("tops Kendall Jenner and I would both like"), refine search results with exact natural language ("this but more sparkly, under 60, and cotton"), virtually try on with your body, and build outfits based on your closet and style.
Rylo is the culmination of the latest in multimodal AI, personalization tech, and years spent studying the most delightful shopping UXs while working in D2C fashion and tech.
- Designing a system that turns a small collection of user-provided photos into hyper-realistic avatars that can try on clothing quickly and efficiently, from a variety of stock photo formats (flat lay, on model, and more), built on a LoRA training pipeline with a strict capture flow of six face angles and two body shots.
- Creating a hybrid search system to capture semantic intent across 100,000+ products, plus a similar-clothing recommendation system, using Marqo embedding models and Pinecone vector search alongside Supabase JSON.
- Scraping 100,000 products from top Shopify stores as the catalog, layering in Google Shopping SERP APIs, and weighting between web search and native catalog search to deliver the strongest results for each query.
- Designing the Rylo AI agent to recommend clothing, search across inventory, and mix and match given multimodal context on the user and what is in their closet: it plans one to four targeted queries from your preferences and chat history, runs them in parallel, and interleaves the ranked results.
- Building and designing the full AI iOS and web application, from onboarding to search, checkout, and many more features.
A SwiftUI iOS client talks to a TypeScript backend that runs both as an Express server on Railway and as a Hono app on Cloudflare Workers, sharing one service layer for near-zero cold starts and global edge distribution. Product data flows in through Serper, SerpAPI, and Oxylabs, auth and storage run on Supabase, and a weekly job plans themed collections for every user from their preferences and recent conversations.
The iOS app.
The web app.
