Slay: Style that Suits You

Inspiration

Everyone has a taste and style in fashion, we are finding your taste and flipping your shopping app experience according to it.

What it does

Our AI-powered e-commerce platform revolutionizes online shopping through personalized styling. By using computer vision and machine learning, the system analyzes clothing styles and learns user preferences. It recommends complementary accessories like belts, watches, and jewelry while considering individual skin tones and fashion tastes. The platform features an intuitive interface where users can swipe to refine recommendations and receive personalized fashion advice from an AI chatbot. This approach directly addresses traditional e-commerce limitations by offering tailored suggestions that increase customer engagement and purchasing likelihood. Research shows personalized recommendations can boost buyer interest by up to 75%, making our AI-driven styling platform a game-changer in online shopping experiences.

How we built it

The application is built on 3 fronts. The AI algorithms, the backend and apparel database and the mobile application frontend. AI

We got a sample flipkart fashion dataset with more than 130,000 items, pre-processed to around 14-15 useful features. We picked up the product images and product descriptions, converted them into CLIP and word embeddings and mapped them to the product and upserted them in a vector database.
Implemented a monte-carlo based swiping mechanism that will suggest similar products if swiped right, dissimilar products if swiped left and swipe up to add to cart.
Made Ava, a fashion-centric advisor, that can answer general queries anything related to outfits, makeup and even accessories. It has access to the Internet, so it's really up with the trends.

Backend

Built a really robust backend with important user, cart and garment schema using PostgreSQL
Made Flask endpoints of all the AI based features that can be called from the front-end easily and made a custom load balancer to route the queries to specific API endpoints.

Frontend

Took inspiration from tinder (hehe) and pinterest mood-boards for the designs and used Flutter to build it.

Challenges we ran into

We wanted the backend (running on a separate system) to connect with the front (running independently) to communicate keeping it under the same subnet and used all our networking knowledge to make it work, didn't really happen. We then did some firewall tweaks and made a work around to make it work in the same system and it worked well.
Was not able to figure out initially how to upsert embeddings in pinecone, that took a really long time to work.

Accomplishments that we're proud of

We were able to build and ship multiple features both GenAI based and traditional AI based in such a short time span
We were able to build a complete product, right from a polished front-end to robust backend embedded with AI features.
Everyone had very specific and unique skill-sets and contributed very well in building it together. ## What we learned
Dont keep the integration for the end, make small tests here and there to check if the methods are working.
Optimising for speed and latency from the ideation phase is a big win.

What's next for Slay: Style that Suits You

We and to add a more detailed search so that user can describe the appearal as accurately as possible and we can deliver the most fitting one. Add a thrift mode, that can be used by both sellers and thrifters alike, making the fashions space more sustainable