-
-
User Profile
-
Chatbot
-
Moodboard Analysis Interface
-
Results of Bad fit
-
Suitable Outfit moodboard and full length image
-
Bad Outfit moodboard and full length image
-
Analysis Page
-
Main Page
-
Rendered results for Suitable Outfit
Inspiration
Fashion inspiration is everywhere, but knowing whether an outfit will actually suit you is much harder. A look can look great on a moodboard or Pinterest, but its colors, silhouette, patterns, and styling may not work equally well for everyone.
We wanted to bridge that gap with Personae, an AI-powered personal stylist that transforms fashion inspiration into personalized recommendations and lets users visualize the result through virtual try-on.
Our goal was to make fashion discovery more personal: instead of simply asking "What is trending?", Personae asks "What works for me?"
What it does
Personae builds a personalized fashion profile by analyzing the user's skin tone, undertone, body shape, and face shape.
Users can then upload a fashion moodboard or outfit inspiration image. Personae:
- Decomposes the moodboard into individual fashion items.
- Identifies properties such as category, color, silhouette, pattern, and material.
- Scores each item against the user's personalized fashion profile.
- Provides personalized recommendations for colors, fitting styles, materials, patterns, and jewelry.
- Handles colors that fall outside predefined recommendation lists using an additional AI color-matching step.
- Selects the best-matching clothing items for virtual try-on.
- Uses the YouCam API to render the selected look on the user's own photo.
The result is an end-to-end experience:
Fashion inspiration → AI analysis → Personalized matching → Virtual try-on
How we built it
Personae is a full-stack application built with React, Vite, Django REST Framework, and Python.
The backend combines several AI approaches:
- Vision LLM for visual profile analysis and moodboard decomposition.
- Fuzzy logic with 60 rules for structured fashion recommendations.
- LLM-based item matching to evaluate individual fashion items.
- A dedicated LLM color-fallback judge for colors not explicitly covered by the recommendation engine.
- YouCam s2s API for virtual try-on and skin-tone color cross-checking.
The complete style-check pipeline runs server-side. After the user's profile is established, the moodboard is analyzed, individual items are scored, suitable items are selected, reference crops are prepared, and the YouCam VTO pipeline renders the resulting look on the user's photo.
We also built explicit handling for different item categories, rendering order, confidence-based item selection, API failures, and pipeline timeouts.
Challenges we ran into
One of our biggest challenges was making fashion matching work beyond simple predefined color categories. Fashion contains countless shades and variations, so an item should not automatically be rejected just because its exact color is not present in a recommendation list.
To solve this, we created a color-fallback system that sends uncovered colors to a dedicated LLM judge for additional evaluation.
The virtual try-on pipeline was another major challenge. Different categories such as clothing, shoes, and bags require different handling, and the order of VTO operations matters. We implemented a chained process with category-specific logic to preserve previously applied garments and the scene.
We also had to handle multiple items targeting the same body region, temporary VTO result URLs, API timeouts, failed renders, and invalid reference images.
Accomplishments that we're proud of
We are especially proud of building a complete AI fashion workflow rather than a standalone recommendation model.
Personae combines computer vision, LLMs, fuzzy logic, fashion reasoning, and virtual try-on into a single user experience.
Some highlights we're particularly proud of:
- A 60-rule fuzzy recommendation engine covering combinations of skin tone, undertone, and body shape.
- AI-powered decomposition of complete fashion moodboards into individual items.
- Item-level matching with an additional color-fallback mechanism.
- A real YouCam-powered virtual try-on pipeline.
- Confidence-based selection and category-aware VTO processing.
- A full-stack application with a polished, fashion-focused interface.
Most importantly, we were able to turn a static fashion inspiration image into a personalized visualization on the user's own photo.
What we learned
This project taught us that building an effective AI product is not only about choosing powerful models. The real challenge is orchestrating multiple AI components into a reliable end-to-end experience.
We learned how deterministic systems such as fuzzy logic can complement generative AI instead of replacing it. Fuzzy rules provide structured and consistent recommendations, while LLMs are useful for visual interpretation and handling ambiguous cases.
We also learned a lot about integrating external APIs into production-style pipelines, particularly around validation, error handling, timeouts, rendering order, and fallback strategies.
Most importantly, we learned how virtual try-on can make AI recommendations much more useful by allowing users to actually visualize the suggested result.
What's next for Personae
The next step for Personae is to expand from a prototype into a more comprehensive AI fashion platform.
We would like to improve personalization over time by incorporating user feedback, preferences, and interaction history. We also want to expand product-level recommendations so that users can move from discovering a suitable style to finding actual products that match it.
Future improvements could include:
- More detailed and continuous fashion profiling.
- Personalized recommendations based on previous user feedback.
- Integration with fashion retailers and product catalogs.
- More advanced virtual try-on capabilities.
- Improved support for accessories and complete styled looks.
- Personalized shopping and wardrobe assistance.
- More robust deployment and scalability for real-world users.
Ultimately, we want Personae to become a personal fashion companion that helps users discover, evaluate, and visualize styles that genuinely feel like their own.
Log in or sign up for Devpost to join the conversation.