Agora

A Multimodal, Agentic Ecosystem Revolutionizing Personalized Fashion through Long-Term Style Memory

Comment

🚀 Project Overview

Agora is not just a chatbot; it is a suite of specialized, multimodal agents that manage the entire fashion lifecycle—from inventory ingestion to personalized trend forecasting and virtual fitting. By leveraging Gemini 3’s multimodal capabilities and Google Search Grounding, the platform creates a "Style DNA" for every user, ensuring that every interaction is informed by past preferences, current trends, and real-time visual data.

🤖 The Agent Fleet

Agent Core Functionality Gemini Power Play
Catalogue Ingestor Converts unstructured PDF catalogues into structured database schemas. Multimodal Extraction: Parses complex layouts and images directly from PDFs.
Personal Stylist Researches trends and conducts interactive user discovery via dynamic UI. Google Search Grounding: Bridges the gap between internal data and global fashion trends.
Matchmaker Pairs the "User Journey" with the product database to curate outfits. Reasoning & Planning: Navigates complex constraints to find the "perfect" match.
Fitting Assistant Provides visual lookbooks and fitting options for selected items. Visual Understanding: Interprets product aesthetics to offer styling advice.
Style DNA (The Brain) Evolves based on OOTD uploads and app interactions; identifies "Celebrity Twins." Long-term Memory: Utilizes a persistent memory mechanism to track style evolution.

🛠️ Technical Innovations

🧠 Persistent Style DNA & Long-term Memory

While most agents reset after a session, our Style DNA Agent maintains a persistent state. It uses a custom memory mechanism to analyze every interaction (and OOTD photo uploads), ensuring the Personal Stylist and Matchmaker agents grow more accurate over time.

🌐 Real-World Grounding

We don't guess what's "in." The Stylist and Style DNA agents utilize Google Search Grounding to reference real-time celebrity styles and runway trends, providing users with cited sources and clickable links for maximum transparency and inspiration.

🎭 Agent2UI: Dynamic Interaction

We’ve moved beyond text. Our agents don't just talk; they act. Using a custom Agent2UI bridge, agents can trigger specific UI components (like preference sliders, color pickers, or lookbook galleries) based on the conversation's context and the agent's unique "personality."

⚡ Transparent Reasoning (Streaming Thoughts)

To build user trust, we expose the "Thinking" process. All agents stream their reasoning steps, allowing users to see how the AI moves from a vague query to a curated fashion recommendation.

A Note on the "Gemini Edge"

This project pushes the boundaries of Gemini by treating it as a reasoning engine rather than a text generator. By utilizing Gemini 3 multimodal windows, we process visual OOTDs, PDF catalogues, and real-time web data simultaneously to provide a truly cohesive styling experience.

Challenges we ran into

Balancing Intelligence with Latency: One of our biggest hurdles was managing rate limits while maintaining a high-quality user experience. Orchestrating five specialized agents via LangGraph meant we hit API quotas quickly.

Accomplishments that we're proud of

  • Redesigning the Shopping Experience: We moved beyond the standard "search bar" paradigm. By combining distinct skill sets—UI/UX design, Agentic AI, and Full-stack Engineering—we successfully built a Dynamic Agent2UI Bridge. We are incredibly proud that our agents don't just chat; they control the interface, summoning sliders and galleries when the context demands it.
  • True Multimodal "Style DNA": We didn't just build a chatbot with memory; we built a system that "sees" style. Successfully integrating user OOTD uploads with long-term memory means Agora understands a user's evolution over time, not just their current session.
  • Transparent AI Reasoning: We implemented a server-sent events (SSE) stream that exposes the agent's "thought process" in real-time. Seeing the Stylist agent actively "Googling" a celebrity trend and then "filtering" for price constraints builds a level of user trust that opaque black-box recommenders can't match.

What's next for Agora

  • From Agent to Marketplace: While we started with fashion, the Agora architecture is domain-agnostic. We plan to scale the Catalogue Ingestor to handle Home Decor, Beauty, and Electronics, turning Agora into a universal, personalized shopping concierge.
  • Social Style DNA & Celebrity Twin Gamification: We plan to expand the Style DNA's ability to identify "Celebrity Twins" into a fully gamified experience, allowing users to instantly match their aesthetic to trending pop culture icons. Additionally, we are introducing a social layer where users can share their "Style DNA" profiles, enabling the Matchmaker agent to automatically curate coordinating outfits for couples or friend groups.
Share this project:

Updates