AI Agentic Upgrade for Online Boutique on GKE

• 

  Architecture Diagram

Inspiration

• I wanted to demonstrate how enterprises can enhance existing microservices without having to rewrite them. Online Boutique is a popular demo app, but it lacked an external AI-powered recommendation service.
• My inspiration was to inject an “agentic” AI layer that can observe, interpret, and augment user experiences without touching the core codebase.

What it does

• The AI Agent microservice analyzes Online Boutique’s product catalog and frontend data, then delivers personalized recommendations via a new /api/recommend/{user} endpoint.
• It supports diagnostics with /api/health/details, scrapes data when JSON APIs fail, and is designed to be extended with Google’s Gemini for reasoning and ranking.

How we built it

• I have built a FastAPI service packaged into a Docker image, deployed on Google Kubernetes Engine (GKE).
• It integrates with Online Boutique via environment variables for product APIs, uses scraping fallback for robustness, and exposes health endpoints for observability.
• The container was built with Google Cloud Build, rolled out via kubectl set image, and tested with kubectl port-forward and ingress routing.

Challenges we ran into

• Dealing with gRPC-only services (catalog) vs HTTP probes.
• Debugging 404 errors on JSON endpoints and falling back to HTML scraping.
• Getting Kubernetes ingress routing and root-paths aligned for the new microservice.
• Managing container rollout delays due to command/args patch mismatches.

Accomplishments that we're proud of

• Delivered a working hosted service on GKE (https://34-54-178-42.nip.io).
• Built robust health checks and scrape-based fallback logic.
• Extended Online Boutique with an “AI-first” agent microservice without touching core services.
• Containerized and automated builds with Cloud Build and GKE deployments.

What we learned

• The importance of designing resilience into microservices (graceful fallbacks).
• How GKE ingress, health checks, and deployments interact in real-world scenarios.
• That scraping can be a practical temporary bridge when structured APIs fail.
• The nuances of fast container rollouts and dependency resolution in slim Python images.

What's next for AI Agentic Upgrade for Online Boutique on GKE

• Enable Gemini-powered ranking for recommendations with richer reasoning.
• Add multi-channel Agent-to-Agent (A2A) flows to support conversations across services.
• Integrate MCP (Model Context Protocol) to provide contextual grounding for recommendations.
• Publish a blog post + LinkedIn demo with #GKEHackathon and #GKETurns10.
• Explore enterprise-scale resilience patterns like service meshes and auto-healing probes.

Team

Solo: Sweety Seelam

Built With

• adk-integration-hooks
• docker
• dotenv
• fastapi
• google-cloud
• google-container-registry
• google-generative-ai
• google-kubernetes-engine
• ingress
• kubectl
• kubernetes-deployments
• python
• requests
• uvicorn