Agentic Online Boutique

Agentic Online Boutique demonstrates AI-powered e-commerce with collaborative agents. Shop through conversation, get design recommendations from photos/text, while AI optimizes pricingautomatically.

Comment

Inspiration

The idea came from noticing how traditional e-commerce makes shoppers wade through endless menus and filters. We wanted to reinvent shopping as something more natural—like chatting with a knowledgeable store assistant. And when shoppers feel bored or just curious, they can prompt the AI to uncover hidden gems—turning the experience into a fun treasure hunt full of delightful surprises!

What it does

Agentic Online Boutique transforms the Google Cloud microservices demo into an AI-powered shopping platform. Users can interact with a Gemini AI shopping agent using natural language queries. The AI understands intent, searches products, manages shopping carts, and provides personalized recommendations by orchestrating multiple backend microservices through a secure Model Context Protocol (MCP) server.

How we built it

We built two core components: a Shopping Agent powered by Google's Gemini 1.5 Flash model that handles natural language understanding and intent recognition, and an MCP Server that provides secure API access to the existing microservices via gRPC. The Shopping Agent runs as a FastAPI service (3 replicas) while the MCP Server (2 replicas) acts as a bridge between AI agents and backend services like product catalog, cart service, and recommendations. We deployed everything on Kubernetes with proper health checks and resource management.

Challenges we ran into

The biggest challenge was integrating AI capabilities with existing microservices without disrupting the original architecture. We had to handle gRPC communication complexities, manage service dependencies, and ensure graceful degradation when services were unavailable. Memory constraints also forced us to exclude the cart service in some deployments, requiring creative workarounds for cart functionality testing.

Accomplishments that we're proud of

We successfully created a production-ready agentic AI system that demonstrates true multi-service orchestration. The Shopping Agent can understand complex natural language queries, coordinate across 8+ microservices, and provide intelligent shopping assistance. We built comprehensive testing tools and a beautiful standalone web interface that showcases the AI capabilities without requiring complex setup.

What we learned

We learned how to effectively bridge AI models with existing microservice architectures using the Model Context Protocol. We gained deep insights into managing AI service deployments on Kubernetes, handling real-time natural language processing at scale, and creating secure API gateways for AI-to-microservice communication. Most importantly, we discovered how to maintain system reliability while adding sophisticated AI capabilities.

What's next for Agentic Online Boutique

Future enhancements include implementing the Phase 2 components we've designed: an A2A (Agent-to-Agent) broker for multi-agent communication, an orchestrator for complex workflow management, and a pricing optimizer agent for dynamic pricing. We also plan to add persistent conversation context, voice interface capabilities, and integration with additional AI models for specialized tasks like visual product search and personalized styling recommendations.

Built With

Share this project:

Updates