PersonalStylist
Taste already exists in the camera roll: outfits you wore, screenshots you saved, and short videos.
Core trick
Argon reads a closet of tagged photos and cites item ids for every taste claim, then checks shopping candidates against that profile.
Description
Taste already exists in the camera roll: outfits you wore, screenshots you saved, and short videos. Shopping search does not know any of that. A recommendation that ignores what you own, what you are dressing for, and what you have already rejected is just another catalog.
Personal Stylist turns that raw material into a closet and a taste profile, then uses both while it shops.
Photos you are wearing become Worn. Screenshots, and photos you add from the Aspirational tab, become Aspirational. Clothes on other people in a normal photo are dropped, so the closet does not fill up with strangers.
One Argon pass reads the whole closet, including the crop images, plus sizes, budget, likes, dislikes, and saved looks. Every claim in the taste profile cites item ids.
The chat agent plans a search, looks at real product pages, and checks each candidate against the profile and the closet before it recommends it. Weak picks are rejected out loud.
Likes, dislikes, and saved looks update the profile. Chat text does not.
Try-on chooses among your own photos (day and evening, city and nature) and uses the occasion you described in chat, such as a hike or an evening gala, when picking the setting.
README excerpt
# Personal Stylist An iPhone stylist that learns what you wear from your camera roll, shops the web for pieces that fit that taste, and shows you in them. The try-on is not a single guess: it renders, watches the result, critiques it, and renders again. ## The hackathon problem The Gemini 4 Argon hackathon asks for work in at least one of two statements. **Problem Statement One: Million Token Engineering.** Argon is strong at long-horizon software work, with a very large output window. The prompt is to take on a monumental engineering task, such as a legacy migration or framework upgrade, and measure the improvement over a baseline. This app is not that kind of project. **Problem Statement Two: More Dimensions.** Argon is also strong at long video understanding, and a single generated draft is only a first pass. The prompt is to build a system that renders what it made, watches the result, critiques it against the goal, and fixes it in a loop. Personal Stylist is built for Problem Statement Two. **See it on me** is that loop, applied to clothing: 1. Nano Banana renders you wearing a product the shopping agent found, using a real photo of you and the product image. 2. Argon watches that render next to the product photo, your photo, and your taste profile. 3. It scores garment fidelity, identity, fit, style match, and an overall score from 0 to 10, and writes concrete edits. 4. The next render applies those edits. This repeats for at most three passes, or stops when eve