OutfitBoss

Inspiration

Despite having a full closet, my girlfriend never knows what to wear. Instead of visiting all her favorite brands, one by one, what if there was a better way to purchase cohesive outfits?

Googling for clothes might be great for finding particular items you have in mind. However, if you want to start from scratch or buy full cohesive outfits for a location or an occasion, our app can help.

What it does

All you have to do is type in what you're looking for, in as much detail as possible. Our agent will then scour our pre-tagged and classified sources to find what goes well together and sends you our recommended outfits.

How we built it

We compiled a directory of clothing outfits, where features from the name, description, and image are extracted. We used a variety of models, including LLaVA (Large Language and Vision Assistant) but in the end settled on DeepSeek. The model selects the most applicable tags from an exhaustive list of tags we generated ourselves. We created our web frontend and backend through NextJS with Shadcn materials. The user prompt is then fed to DeepSeek which picks the tags from our exhaustive list that match the best.

Challenges we ran into

Most models were terrible at image feature extraction. We tried a bunch and settled on DeepSeek due to time constraints. Scraping a large number of outfits from various outlets was a time consuming process. Each outfit also had to go through our feature extraction process.

Accomplishments that we're proud of

Our frontend looks great and it's just a matter of increasing the number of vendors and listings in our directory and optimizing the tag generation process.

What we learned

Choice of model is very important. We also learnt a lot about UI design from our frontend. Learnt how to use LMStudio to locally host LLMs. Learnt how to use MongoDB Atlas. Learnt how to connect everything in NextJS and Typescript.

What's next for OutfitBoss

Use a different machine learning model (not an LLM) for image feature extraction. Keep scraping outlets or find a different way to find products.

Built With

• deepseek
• farfetch
• llava
• lmstudio
• mongodb
• nextjs
• nike
• selenium
• typescript
• uniqlo