SmartWardrobe AI

Inspiration

Standard large language models completely lack native regional climate context and cannot natively respect individual user style or modesty boundaries. We built an intelligent assistant to bridge this gap automatically.

What it does

SmartWardrobe AI instantly fetches live regional weather metrics via the Model Context Protocol (MCP) based on a user's location. Users can upload outfit photos, which are automatically scanned by Google Gemini 2.5 Flash vision intelligence to detect specific design cuts (like sleeveless, cropped, or deep cuts). If a personal style or modesty preference is flagged, a programmatic guardrail engine instantly provides real-time layering fixes, such as suggesting a structured shrug or linen blazer, ensuring the ensemble matches both the local climate and personal comfort boundaries.

How we built it

We engineered a multi-agent orchestration architecture using Python and Streamlit for the web platform. The system leverages the Model Context Protocol (MCP) via an asynchronous local tool server to fetch real-time climate data. For visual parsing, we integrated Google Gemini 2.5 Flash via the official SDK, passing raw image data alongside strict, multi-shot structured prompts to guarantee deterministic tag outputs. Finally, we built a rule-based algorithmic verification layer to instantly evaluate those attributes against custom style boundaries and trigger dynamic fallback suggestions.

Challenges we ran into

Integrating asynchronous FastMCP tool functions with Streamlit's real-time interface required careful state handling to prevent UI freezing. Prompting the vision model to consistently return predictable text tags from diverse clothing photos without extra conversational fluff required extensive prompt tuning. Additionally, overriding Streamlit's automatic text box caching when switching locations rapidly required engineering a custom key-reset mechanism.

Accomplishments that we're proud of

We successfully built and deployed a fully functional, end-to-end multimodal AI platform within hours. We are proud of seamlessly integrating advanced computer vision with the new Model Context Protocol standard to solve a real-world problem, delivering live regional climate data and intelligent, preference-aware style guardrails in a clean web application.

What we learned

We learned how to design and deploy specialized agentic workflows that bridge real-world tools with large language models. Specifically, we gained deep experience in handling multimodal image payloads with the Gemini API, working with the new Model Context Protocol (MCP) standard to fetch external context, and managing state and deployment logic in cloud-hosted Streamlit environments.

What's next for SmartWardrobe AI

The next step is expanding the system into a complete smart closet ecosystem. We plan to integrate a vector database to match uploaded photos against live, real-time e-commerce catalog inventories, allowing the AI to recommend actual purchasable modest alternatives. Additionally, we aim to introduce deep personalization layers that learn a user's style history over time and add support for multi-garment outfit pairing suggestions.

Built With

• fastmcp
• google-gemini-2.5-flash
• model-context-protocol-(mcp)
• pillow
• python
• streamlit