Sensibility

Senibility

Inspiration

Fashion is a powerful form of self-expression, but choosing what to wear isn’t always easy. There are moments when you want to present yourself differently, whether it’s dressing up for a networking event, going out with friends, attending a party, or simply matching a new vibe or aesthetic. Even with a full closet, decision fatigue and uncertainty often get in the way.

We built Sensibility to help people confidently express their identity through fashion by making outfit decisions effortless, personalized, and adaptive over time.

What it does

Sensibility is an AI-powered fashion assistant that recommends outfits tailored to a user’s personal style and aesthetic.

Users can:

Take and upload photos of their own clothing
Choose or describe an aesthetic or occasion
Receive outfit recommendations using items they already own
Rate outfits so the system learns and improves over time
User page gives descriptive analysis of their preferences to their style, what they dislike and like, and the AI adapts to your style
Gives Users wardrobe gap analysis

The more users interact with Sensibility, the better it understands their preferences, creating increasingly personalized recommendations.

How we built it

We built Sensibility as a full-stack, mobile-first application using:

Frontend: React Native with Expo (TypeScript)
Backend: FastAPI (Python)
Computer Vision: CLIP (for vision-language similarity and style matching), BLIP (for image captioning and garment understanding)
Large Language Models: OpenAI GPT-4o-mini (style labelling, explanations, recommendations)
Anthropic Claude 3 Haiku (behavioural pattern and preference analysis)
Machine Learning: PyTorch
AI Orchestration & Memory: Backboard.io for multi-model switching and adaptive memory
Created REST APIs to integrate the backend with the frontend
Supabase (for database)

This multi-model pipeline allows us to select the best AI model for each task, from image understanding to behavioural analysis.

Challenges we ran into

Learning how the CLIP model works and how to effectively use vision-language embeddings
Building a React Native app for the first time
Working with computer vision, which was new to our team
Integrating the frontend and backend, including handling network and URL configuration issues
Understanding and properly implementing Backboard.io for model switching and persistent memory

Despite these challenges, we successfully built a functioning end-to-end product.

Accomplishments that we're proud of

Building a fully functional mobile app from scratch
Successfully integrating multiple AI models into a single pipeline
Implementing adaptive memory so the system learns from user interactions
Learning and using Backboard.io effectively for AI orchestration
Creating a real, usable product that improves with every interaction

What we learned

React Native and mobile app development
Computer vision fundamentals with CLIP and BLIP
Multi-model AI orchestration using Backboard.io
Designing AI systems that learn from behavioural data, not just static inputs
Bridging frontend UX with backend AI systems

What's next for Sensibility

Next, we plan to expand Sensibility with deeper style analytics and long-term style evolution tracking. Our goal is for Sensibility to not just recommend outfits, but to grow alongside the user, understanding how their identity, preferences, and style change over time.

Future features include: