Building Revolgy’s Google Cloud Summit Assistant with Google ADK

At Revolgy, we’re passionate about showcasing how AI can transform customer experiences, especially at high-impact events like the Google Cloud Summit. Our Summit Assistant is a multi-agent, chat-based tool built with Google’s Agent Development Kit (ADK) to guide attendees through agendas, speakers, and real-time Google Cloud documentation. Here’s a look under the hood at how we brought this live demo to life.

1. Multi-Agent Architecture

We designed the backend as a collection of specialized agents, each with a focused skill set:

• Summit Agent reads structured summit data (from summit_details.json) and answers questions about sessions, times, and speakers.
• GCP Docs Agent, GCP Product Releases Agent, and GCP APIs Agent leverage Vertex AI Search datastores to provide authoritative answers on docs, product releases, and APIs.
• Finally, Google Search Agent falls back to standard web search when needed, adding a playful “fun fact from 10 years ago” to each reply for extra engagement.

All agents are orchestrated by a root agent in a FastAPI application, with state managed in an in-memory session store. The runner thread invokes the right sub-agent based on user intent and merges results into a unified chat response.

2. Backend Service & Deployment

• Framework: FastAPI powered by Uvicorn, wrapped by ADK’s helper function, makes it easy to expose agent endpoints.
• Data Storage: Summit agenda JSON lives alongside the code; environment variables configure Vertex AI datastore IDs.
• Containerization: The application is containerized for easy deployment into Cloud Run.
• Cloud Run: We deploy the backend with environment variables for GCP project, region, and datastore IDs, enabling zero-maintenance scaling during the summit.

3. Streamlit Frontend

For the user interface, we chose Streamlit:

1. Chat Experience: A clean chat layout reads/writes messages to FastAPI via REST calls.
2. Persistent State: Session IDs and history are there to enable users to ask follow-ups without losing context.
3. Configuration: An API_URL environment var points to the deployed backend.

The result is a responsive, single-page chat app that can run locally or in Cloud Run with minimal tweaks.

4. What This Demonstrates

By building and deploying this Summit Assistant, Revolgy illustrates how easy and fast you can deploy conversational AI with ADK’s high-level abstractions. Easy integration of custom data (event agenda) with enterprise-grade search (Vertex AI Search) allows customization of your application. By deploying this in Cloud Run, you can take advantage of serverless compute with scalable deployment using container best practices.

Finally, the engaging UX via Streamlit enables non-developers to spin up a chat interface in minutes.

5. Getting Started

1. Clone the repository[link-to-repository].
2. Set your environment variables.
3. Run the frontend:

   cd frontend
   pip install -r requirements.txt
   streamlit run app.py

4. Run the backend:
   

   cd backend/summit-app
   pip install -r requirements.txt
   adk api_server

5. Point your frontend at http://localhost:8000 and start chatting about Summit Prague 2025!

Revolgy’s Summit Assistant is a live testament to how quickly you can build a polished AI tool with the Google Agent Development Kit, which is perfect for events, demos or customer support. We can’t wait for you to play with it at the Google Cloud Summit!

 We prepared AI Adoption program for teams or managers who:

• are interested in AI but not sure where to start
• need advanced guidance in AI usage or building
• have special use-cases that they need to solve
• want to learn about AI and need training and workshops

Contact Revolgy today to find out how we can help you get started.