Pocket Closet
From Product Designer to AI-Native Builder: how I used AI to design, build, and launch a Gemini-powered web app I now use every day to get dressed in the morning.
Platform
Vibecoded Web App
Info
Web App / Cursor / Vibe Coding / Gemini API
Timeline
4 weeks
I vibe coded a web app to solve a common daily problem: waking up in the morning with a full closet, but somehow having nothing to wear.
With AI was my build partner, I designed and launched a product from the ground up the gives you AI-powered outfit suggestions using clothing from your own closet.
Live demo in Cursor
Your closet in your pocket: how it works
Users start by uploading photos of their clothing, and Gemini creates beautifully accurate virtual 3D renditions, analyzing and assigning metadata and organizing your own virtual catalogue. Then, Gemini provides thoughtfully curated outfit suggestions based on customizable filters, open-ended input prompts, and an optional weather widget to take the day's weather conditions into account.
Live demo Image generation with Gemini
Turning messy photos into a curated virtual closet
To start, I worked with AI in Cursor to shape the behind-the-scenes prompting structure that turns messy phone photos into a clean studio shots. When users upload clothing photos, Gemini turns messy backgrounds and wrinkles into crisp studio images of the user's clothing. At the same time, it analyzes the item and tags metadata, turning every piece into structured data and building a map it can interperet later for outfit curation.
Live demo outfit suggestions with Gemini
Outfit suggestions for every occasion
The home feed works like a personal stylist. It builds full outfits from your own pieces and factors in the weather, the occasion, and whatever you type in, like "I want to wear pink today." Users can favorite the best looks for later, and if you thumbs-down something it won't show up again.
Design System Visual Style Guide
Consistency in code: Building an AI-native design system
I worked with AI as a build partner, providing examples of how I wanted the design system to look and feel with eferences and moods I was going for, so it understood my intent before writing any code. From there I built the whole design system from the ground up in Cursor as a single source of truth in code, for the components, colors, type, spacing, motion curves, and interaction patterns, with a visual style guide to edit and reference.
Screens from the model style evaluation test
Evaluating AI models for style curation
I wanted to see which model had the best sense of style, so I worked with AI in Cursor to design a blind "taste test" pitting Gemini's fashion taste against Claude's. Hooking in both APIs, I asked each model to generate outfits from my virtual closet side-by-side using the same prompts, with a data-driven read out telling me which model's style curation I preferred.
