-
-
Logo
-
Chat Agent
-
Live Agent
-
Dashboard
-
Profile
-
Closet
-
Combines
Inspiration
We all share a common paradox: a closet full of clothes but "nothing to wear." We realized that the problem isn't a lack of clothes, but a lack of intelligent, proactive guidance. We wanted to solve this decision fatigue by creating an autonomous agent that knows your style better than you do and can show you exactly how an outfit will look before you put it on.
The launch of Gemini 3 and the Action Era inspired us to rethink fashion assistance. Instead of building another chatbot that answers questions, we decided to build Comby: an autonomous fashion agent that orchestrates multiple AI capabilities to manage your entire style journey—from real-time outfit analysis to proactive travel packing.
What it does
Comby is an agentic fashion orchestrator powered by Gemini 3, designed to proactively manage your entire style journey. Unlike traditional fashion apps, Comby doesn't just respond to queries—it anticipates your needs, adapts to changing conditions, and acts autonomously.
• Live Style Guardian (Real-Time Analysis): Using Gemini 3's Live API, Comby instantly analyzes your outfit through your camera, providing immediate feedback on weather compatibility, color coordination, and occasion appropriateness. The agent reasons over video in real-time, delivering context-aware insights without latency.
• Multimodal Reasoning: Upload any inspiration image (e.g., Justin Bieber's style, Pinterest boards), and Comby's multimodal reasoning instantly matches it with your existing wardrobe. It analyzes the image, identifies missing pieces, suggests alternatives from your closet, and generates outfit combinations—all by reasoning over image + text + structured wardrobe data simultaneously.
• Agentic Shopping Integration: When Comby identifies missing wardrobe items, it autonomously triggers Google Shopping to discover perfect matches. The agent doesn't wait for user approval; it proactively finds options and presents them as part of the styling solution, seamlessly bridging the gap between styling and purchasing.
• Marathon Agent (Long-Running Task Management): Planning a trip? Comby's Marathon Agent proactively monitors your calendar, analyzes weather forecasts for your destination, and generates travel outfit suggestions. If weather changes mid-week, the agent self-corrects and sends proactive notifications—updating recommendations without you asking.
• Creative Autopilot with Virtual Try-On: Comby visualizes recommended outfits on you using Nano Banana Pro, delivering photorealistic virtual try-on experiences. Users see exactly how an outfit will look before committing, reducing purchase hesitation and decision fatigue.
• Self-Correction & Privacy-First Design: When location permissions are denied, Comby politely asks for your city and continues its reasoning. It never breaks the user experience; it adapts. This demonstrates Self-Correction in action—a hallmark of true agentic behavior.
How we built it
• Gemini 3 Live API Integration: We leveraged Gemini 3's Live API to process video streams in real-time. The agent analyzes outfit components, weather context, and occasion fit simultaneously, delivering instant feedback without round-trip latency. This enables true real-time style guidance.
• Multimodal Reasoning Architecture: Gemini 3's native multimodal capabilities enable Comby to reason over video input (from Live API), image inputs (inspiration matching), text inputs (user context), and structured data (calendar, weather, wardrobe inventory) all in a single reasoning step. This unified approach enables context-aware, coherent styling decisions that traditional systems cannot achieve.
• Agentic Orchestration & Tool Coordination: We engineered Comby to autonomously orchestrate multiple tools and APIs:
• Google Shopping API: For autonomous product discovery when items are missing
• Google Calendar API: For proactive event-based outfit planning
• Weather APIs: For real-time weather integration and proactive notifications
• Nano Banana Pro: For photorealistic virtual try-on generation
• Vibe Engineering: For autonomous testing and verification of outfit combinations
• Prompt Engineering for Autonomous Decision-Making: We didn't just fine-tune prompts for better responses; we engineered prompts that enable autonomous decision-making. Comby's prompts include:
• Context awareness (weather, calendar, location, user preferences)
• Proactive reasoning (anticipating user needs before they're explicitly stated)
• Tool orchestration (knowing when to trigger Shopping, Calendar, or Weather APIs)
• Self-correction logic (fallback strategies when APIs fail or constraints change)
• Thought Signatures (transparent reasoning steps that users can see and understand)
• State Management & Wardrobe Synchronization: Handling the complex state of a user's digital closet and synchronizing it with real-time data (weather, calendar, shopping options) required careful architecture. We built systems to maintain consistency across multiple data sources while ensuring fast response times.
Challenges we ran into
• Real-Time Multimodal Processing: Getting Gemini 3's Live API to process video streams consistently and deliver instant feedback required extensive optimization. We had to balance latency with reasoning quality, ensuring the agent provides useful insights without delay.
• Multimodal Reasoning Consistency: Ensuring Comby correctly matches uploaded inspiration images with user wardrobes while generating coherent outfit suggestions required sophisticated prompt engineering. We implemented validation layers to ensure outfit suggestions are realistic and wearable, using Thought Signatures to debug and improve reasoning chains.
• Autonomous Tool Orchestration: Coordinating Gemini 3, Google Shopping, Calendar, Weather, and Nano Banana Pro into a seamless workflow required careful state management and error handling. We implemented a robust orchestration layer that manages API call sequences, implements fallback strategies, caches results to minimize latency, and handles rate limits gracefully.
• Achieving True Agentic Behavior: The hackathon explicitly states: "If a single prompt can solve it, it is not an application." We had to ensure Comby wasn't just a chatbot but a true autonomous agent. This required building multi-step reasoning, proactive notifications, long-running task management, and self-correction logic into the core architecture.
• Privacy & User Context: Handling denied location permissions without breaking the user experience was critical. We implemented graceful degradation—when location is denied, Comby asks for the city manually and continues reasoning seamlessly, demonstrating Self-Correction in action.
• Latency & User Experience: Multiple API calls (Gemini, Shopping, Weather, Try-on) can create noticeable delays. We parallelized independent API calls, implemented streaming responses for real-time feedback, optimized prompt complexity to reduce inference time, and added engaging loading states and progress indicators.
Accomplishments that we're proud of
• True Agentic Architecture: We built a system that doesn't just respond to prompts but proactively manages a user's style journey, demonstrating the Action Era vision. Comby breaks down complex tasks into sub-tasks, monitors conditions continuously, and acts autonomously.
• Seamless Multimodal Integration: Comby flawlessly combines video (Live API), images (inspiration matching), text (user context), and structured data (calendar, weather) into coherent styling decisions. This unified multimodal reasoning is unprecedented in fashion technology.
• Autonomous Tool Orchestration: The system independently decides when to trigger Shopping, Calendar, Weather, or Try-on features based on context—no explicit user commands needed. This demonstrates true agentic behavior.
• Transparent AI Reasoning: Users see Comby's thought process (Thought Signatures) in the chat interface, building trust and enabling better human-AI collaboration. This transparency is crucial for user confidence in autonomous systems.
• End-to-End Fashion Solution: From real-time outfit analysis to proactive travel planning to photorealistic try-on, Comby covers the entire fashion decision-making workflow. The system handles everything from inspiration to purchase to visualization.
• Photorealistic Virtual Try-On: Using Nano Banana Pro, Comby generates photorealistic virtual try-on experiences that significantly reduce purchase hesitation and decision fatigue. Users can visualize exactly how an outfit will look on them before committing.
What we learned
• Agentic AI Requires Orchestration, Not Just Prompting: Building a true agent means coordinating multiple systems, managing state, handling failures, and making autonomous decisions. It's fundamentally different from building a chatbot. The complexity lies not in individual components but in their orchestration.
• Multimodal Reasoning is Powerful but Requires Careful Engineering: Gemini 3's ability to reason over multiple modalities simultaneously is transformative, but guiding that reasoning toward consistent, useful outputs requires sophisticated prompt engineering, validation layers, and continuous improvement through Thought Signatures.
• Proactivity Changes User Expectations: Users expect agents to anticipate needs and act without being asked. This shifts the entire UX paradigm from "ask and answer" to "monitor and notify." Proactive notifications and self-correction become essential features, not optional enhancements.
• The 1M Token Context Window is a Game-Changer: Gemini 3's 1M token context window enables maintaining rich user context (wardrobe history, preferences, past interactions, weather data, calendar events) without complex retrieval systems. This simplifies architecture and dramatically improves reasoning quality.
• Self-Correction is Essential for Robustness: Real-world systems fail—APIs go down, permissions are denied, data is incomplete. Building agents that gracefully degrade and adapt (like Comby asking for city when location is denied) is crucial for production-grade applications. Self-correction isn't a nice-to-have; it's essential.
• Real-Time Feedback Matters: Users expect instant responses, especially for real-time features like Live Style Guardian. Latency directly impacts perceived intelligence and usefulness. Streaming responses and optimized API calls are critical for user satisfaction.
What's next for Comby: Your Agentic Closet Companion
• Wardrobe Analytics: Insights into your most worn items, color usage patterns, seasonal trends, and style evolution. The agent will learn from your choices and provide personalized recommendations based on historical data.
• Marketplace Integration: Seamless shopping experience with affiliate partnerships. When Comby suggests items, users can purchase directly within the app, creating a frictionless path from inspiration to ownership.
• Calendar Integration: Proactive outfit planning for upcoming events. Comby will monitor your calendar, analyze event types, and generate outfit suggestions weeks in advance, ensuring you're always prepared.
• Social Sharing: Share outfit recommendations with friends and get feedback. Build a community around style, enabling users to collaborate and inspire each other.
• Seasonal Planning: Proactive wardrobe refresh suggestions based on season changes. The agent will identify gaps in your wardrobe and recommend items to purchase for upcoming seasons.
• Sustainability Tracking: Carbon footprint of outfit choices and recommendations for sustainable alternatives. As fashion becomes more environmentally conscious, Comby will help users make eco-friendly choices.
• Predictive Fashion: Using historical data to predict future style preferences and proactively suggest items. The agent will anticipate your needs and present options before you even realize you need them.
• AR Try-On: Advanced augmented reality for in-store try-on experiences. Comby will extend beyond virtual try-on to enable real-world AR experiences in physical stores.
• Fashion Futures: Collaborate with brands to predict and influence fashion trends. Comby's aggregated data will provide insights into emerging trends, helping brands stay ahead of the curve.
Log in or sign up for Devpost to join the conversation.