BKK — Persona-Based AI Outfit Recommendation System

Persona-driven outfit recommendations combining CLIP retrieval with Gemini multimodal reasoning

What it does#

A team senior design project that recommends complete outfits — topwear, bottomwear, footwear, and accessories — based on a user's style keywords, a persona, and an occasion-driven "combine mode."

Retrieval#

A fine-tuned CLIP (fashion-clip) model embeds garment text and images, indexed in Pinecone for similarity search across a large fashion product catalog, filtered by category and gender.

Outfit selection#

For each garment category, Google Gemini's multimodal model reasons over the retrieved candidates alongside the pieces already chosen, picking the option that best fits the persona, style keywords, and outfit context.

Demo#

A Gradio app lets users enter style keywords, pick a persona and combine mode, and view the generated outfit laid out across all categories, with the model's top pick highlighted for each.