LibaasAI

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

Inspiration We were inspired by how difficult it can be to find clothing items seen on social media, screenshots, or in real life. Traditional shopping platforms rely heavily on text searches, which often fail when users do not know the brand or product name. We wanted to create a smarter and more intuitive fashion discovery experience using AI and computer vision. What it does LibaasAI allows users to upload an image or screenshot of a clothing item and instantly receive visually similar fashion products. The platform analyzes the image using AI models to understand clothing type, colors, style, and aesthetics, then matches it against a product database. Users can browse recommendations, compare prices, and explore similar styles through a clean and responsive interface. How we built it We built the frontend using Next.js and Tailwind CSS to create a fast and modern user experience. The backend was developed with Flask to handle image uploads, API requests, and AI processing. For the AI pipeline, we combined Gemini Vision for clothing analysis and CLIP embeddings for visual similarity matching. Gemini extracts semantic information such as style, fit, and color palette, while CLIP converts images into embedding vectors for similarity search. We then ranked products using cosine similarity: \cos(\theta)=\frac{A\cdot B}{|A||B|} We used a mock fashion dataset for the MVP and connected product metadata such as prices, ratings, and categories to the matching system. Challenges we ran into One of the biggest challenges was improving recommendation accuracy. Pure visual similarity sometimes returned items that looked similar but belonged to completely different fashion styles. To solve this, we combined embeddings with semantic filtering using Gemini-generated metadata. Another challenge was handling noisy images with complex backgrounds, lighting, or multiple clothing items. We explored background removal and clothing isolation techniques to improve matching quality. We also faced challenges integrating AI pipelines with real-time frontend interactions while keeping response times fast enough for a smooth user experience. Accomplishments that we're proud of We are proud of building a fully functional AI-powered fashion discovery prototype that combines modern frontend development with machine learning and computer vision. We successfully integrated image understanding, similarity matching, ranking systems, and a polished UI into one cohesive platform. We are also proud of creating a scalable architecture that can later support larger product databases, smarter recommendations, and advanced fashion features. What we learned Through this project, we learned how AI embeddings work for similarity search and how semantic understanding can improve recommendation systems. We gained hands-on experience integrating computer vision models into a full-stack web application and learned more about API design, image processing, and frontend-backend communication. We also learned the importance of balancing AI accuracy, scalability, and user experience in real-world applications. What's next Our next steps include expanding the product database, improving recommendation accuracy, and adding multi-item outfit detection. We also want to introduce features such as “shop the look,” personalized recommendations, favorites, and real-time price comparisons across multiple retailers. In the future, we hope to evolve LibaasAI into a fully personalized AI fashion assistant that makes online fashion discovery smarter, faster, and more interactive.