An autonomous, multi-agent AI system built for the GKE Turns 10 Hackathon that supercharges the Online Boutique microservices app with real-time marketing, recommendations, and support.
Our project is an external "AI brain" for the Online Boutique. It uses a team of specialized agents that work together to monitor the application, find insights, and take autonomous action without touching the original code.
[User Traffic] -> [Online Boutique] -> [MCP Server] -> [Analyst Agent] --(A2A)--> [Marketing Agent] -> [Database]
| Component | Role in System | Key Technologies Demonstrated |
|---|---|---|
| MCP Toolbox Server | Provides a clean API for real-time user activity context. | GKE, MCP Server |
| Business Analyst Agent | Senses trends using the MCP Server and delegates tasks. | GKE, Gemini, MCP Client, A2A (Sender) |
| Marketing Agent | Receives goals, generates ad copy with Gemini, and publishes it. | GKE, Gemini, A2A (Receiver), Cloud SQL |
| Recommendation Agent | Replaces the original service with an AI-powered gRPC server. | GKE, Gemini, gRPC, MCP (Client), A2A (Receiver) |
| Customer Support Agent | Acts as a user-facing chatbot grounded in the app's data. | GKE, Gemini, gRPC, MCP (Client) |
| Mission Control Dashboard | A live UI that visualizes the entire system's activity. | GKE, Flask |
- Fully Autonomous Closed Loop: Our system autonomously senses real-time events, thinks about their strategic importance, and acts to create a business outcome without human intervention.
- Diverse Agent Patterns: We demonstrate multiple agent archetypes: a scheduled analyst, a reactive marketing service, an intelligent gRPC replacement, and an interactive user-facing chatbot.
- Responsible AI (Grounding): Our Customer Support Agent is explicitly "grounded" in the application's real-time data to provide trustworthy answers and prevent AI hallucination.