Authors

Andriana Filatova, Oleksandr Fer, Marharyta Biloshytska

Abstract

Closy is an AI-powered personal-styling and wardrobe-management project intended to helpusers assemble outfits from garments they already own. It addresses the recurring difficulty ofchoosing what to wear despite having a full wardrobe, a process the team associated with losttime, frustration, decision fatigue, and unnecessary purchases. The system combineswardrobe contents with personal style, weather, occasion, and time of day to producecontextual suggestions. Its stated audiences include students, office workers, busyprofessionals, and others seeking a simpler routine and more sustainable use of existingclothing.

The presented artifact is a functional web-based prototype designed to reproduce the intendedsmartphone experience; the team explicitly distinguished it from a completed mobileapplication. Closy is centred on assembling looks from clothes already owned rather thandirecting users toward purchases. A user can upload a personal photograph, choose a garmentfrom the prototype catalogue, or upload an image of a specific item. Recommendation contextcan be refined by city, occasion, and preferred clothing style.

The interface contains four principal pages. The main page displays a live weather forecastand offers recommendations for the day. The Closy AI page brings together image submission,virtual try-on, contextual controls, and conversational fashion advice. The wardrobe pagestores and organises uploaded items, while the profile page contains user information andapplication settings. In Q&A, the team explained that users photograph their own garmentsand add them to the wardrobe; those images can subsequently inform personalised outfitcombinations. The chatbot was described as able to report weather, search the internet in realtime, and draw upon wardrobe items when composing complete outfits.
After the main action is triggered, a local AI pipeline uses the person and garment images toproduce a try-on preview. GPU memory is then cleared before an image-aware languagemodel performs style analysis and places advice in the chat. A further image model prepareshigh-quality imagery for outfit cards. Weather data is fetched from an external service and usedto vary suggestions dynamically. This staged workflow demonstrates the core interaction andprocessing concepts while retaining a web interface appropriate to the competition deadline.

license

Creative Commons BY-SA 4.0