SoGent

SoGent is an intelligent multi-agent customer assistant, powered by Google ADK and Gemini, to handle complex customer inquiries, provide personalized support, and proactively engage with customers.

Comment

Inspiration

The inspiration for SoGent arose from the growing need for businesses and organizations to deliver fast, intelligent, and personalized customer support in a digital-first world. As customers increasingly expect instant answers and seamless experiences, traditional support channels, such as static FAQs, slow email responses, or basic chatbots, often fall short in providing the depth, accuracy, and engagement that modern users demand.

SoGent was conceived to bridge this gap by leveraging the latest advancements in AI, particularly Google’s Gemini LLM and the Agent Development Kit (ADK), to create a multi-agent platform that can:

  • Handle Complex Queries: Go beyond simple scripted responses by understanding context, researching products, and providing detailed, structured answers.
  • Offer Personalized Support: Use real website data and user context to deliver tailored assistance, product recommendations, and comparisons.
  • Enable Proactive Engagement: Empower support teams with tools to anticipate customer needs and resolve issues before they escalate.
  • Modernize the Support Experience: Deliver a beautiful, intuitive, and interactive UI that feels as responsive and helpful as a real human agent.

The project is inspired by the vision of making advanced AI support accessible to any website or business, regardless of size or technical expertise. By combining modular agents (for greetings, product info, support, and comparison) with a simple, extensible backend and a delightful frontend, SoGent aims to set a new standard for customer service and engagement in the AI era.

What it does

SoGent is an intelligent, multi-agent customer service and engagement platform designed to transform how customers seamlessly navigate their personalized support for any business services/products. Powered by Google ADK and Gemini, SoGent provides a seamless, modern support experience through the following capabilities:

  • Multi-Agent Chat Support: SoGent orchestrates specialized agents for greetings, product information, website support, and product comparison. Each agent is optimized for its domain, ensuring users get accurate and context-aware responses.
  • Website-Aware Assistance: Users can register any website, and SoGent will fetch its metadata. All support and product queries are answered in the context of the selected website, making responses highly relevant.
  • Product Information & Research: Users can ask about any product, and SoGent’s Product Agent will research and return structured details such as product name, description, image, price, and purchase link.
  • Product Comparison: Users can compare two products, and SoGent’s Compare Agent will deliver a professional, structured comparison, complete with feature tables, prices, and a summary, directly in the chat.
  • Proactive & Personalized Support: The SoGent’s Support Agent can answer website-specific questions, guide users through site features, and provide tailored help based on the current context.

How we built it

SoGent was engineered as a full-stack, modular AI support platform, combining modern web technologies with advanced AI and agent orchestration. Here’s how we built it:

  • 1. Multi-Agent Architecture with Google ADK & Gemini: We leveraged Google’s Agent Development Kit (ADK) to define and orchestrate specialized agents: Greeting, Product, Support, and Compare Product Agents. Each agent is powered by Google Gemini (via the Python google-genai SDK), enabling deep language understanding, structured output, and real-time research capabilities. The agents are composed as sub-agents under a root agent, allowing for seamless delegation and teamwork.
  • 2. FastAPI Backend: The backend is built with FastAPI, providing a robust, async API for chat, website management, and agent orchestration.
  • 3. Modern Frontend with Responsive UI: The frontend is a single-page application built with vanilla JavaScript, HTML5, and CSS3 for a sleek, animated chat window that supports text, product cards, and comparison tables.

Challenges we ran into

Building SoGent as a multi-agent, AI-powered support platform presented several unique challenges:

  • Structured AI Output: Getting consistent, structured JSON responses from large language models (LLMs) like Gemini was challenging. The model sometimes returned plain text or malformed JSON, especially for complex outputs like product comparisons. We addressed this by using Pydantic schemas, strict prompt engineering, and the response_schema feature in google-genai to enforce output formats.
  • Real-Time Website Data Extraction: Automatically fetching website contents for contextual consistency is challenging. Some websites lacked standard metadata or used non-traditional icon paths, so we implemented multiple fallback strategies and base64 encoding for reliable display.
  • Professional UI Rendering: Displaying structured AI responses (like product cards and comparison tables) in a modern, responsive chat interface was non-trivial. We iterated on the frontend design and CSS to ensure clarity, accessibility, and a consistent look across devices.

Accomplishments that we're proud of

  • Seamlessly integrated Google Gemini and ADK to create a robust multi-agent support system.
  • Achieved reliable, structured AI responses for both product info and product comparison.
  • Enabled real-time website-aware support and dynamic product research.
  • Designed an extensible architecture that makes it easy to add new agents and features.

What we learned

  • The importance of clear agent roles and prompt engineering to get consistent, high-quality AI responses.
  • How to leverage Google ADK and Gemini’s latest features, including structured output with Pydantic schemas and response schemas.
  • The value of robust error handling and fallback strategies when dealing with real-world website data and unpredictable AI outputs.
  • That seamless frontend-backend integration is key for maintaining context and delivering a smooth user experience.
  • How thoughtful UI/UX design can make complex AI interactions feel intuitive and engaging for end users.
  • The benefits of modular, extensible architecture for scaling and evolving an AI-powered support platform.

What's next for SoGent

  • Advanced Agent Capabilities: Expand the agent ecosystem with specialized agents for order tracking, returns, and personalized recommendations.
  • Multi-language Support: Enable multilingual conversations to serve a global user base.
  • Voice & Omnichannel Support: Add voice chat and support for messaging platforms (WhatsApp, Messenger, etc.) to reach users wherever they are.
  • Continuous Learning: Implement feedback loops so agents can learn from real interactions and improve over time.
  • Security & Compliance: Enhance data privacy, user authentication, and compliance features for enterprise readiness.

Built With

Share this project:

Updates