GreenOps AI Agent

Inspiration

Cloud sustainability is becoming one of the top priorities for modern organizations. While GKE makes running workloads easier, it’s still hard for teams to visualize the real environmental cost of their deployments. We wanted to create a tool that not only shows carbon footprint & costs, and potential enhancement opportunity to add a self-heal agent in future

What it does

Deploys alongside the Online Boutique microservice app on GKE.

• Provides a Streamlit dashboard that: Visualizes GKE costs from BigQuery billing export. Estimates CO₂ emissions using the Carbon Footprint API (with mock fallback). Calls Vertex AI Gemini for recommendations

How we built it

Infrastructure: GKE Autopilot + Online Boutique deployed in boutique namespace. Data: Billing export → BigQuery → GreenOps Agent. Dashboard: Python + Streamlit app for visualization. AI Agent: Kubernetes Python client + Vertex AI Gemini API for explanations and auto-remediation. Deployment: Docker + Cloud Build + GKE LoadBalancer service.

Challenges I ran into

Billing export to BigQuery tio test initially (we added a mock data for instant testing than mapped to live data). Ensuring autopilot clusters had external IPs exposed. Debugging Streamlit + GKE deployment (CORS, port configs). Corbonfootprint API, takes 21 days to reflect (default) Vertex API end points depreciated, so then used Gemini flash model

Accomplishments that we're proud of

Combined sustainability + AI remediation in one dashboard. Made it beginner-friendly (easy to reproduce in Cloud Shell). Designed it so it could scale for enterprise FinOps + GreenOps use cases.

What we learned

How to connect GCP billing + BigQuery + Carbon Footprint API. How to use Gemini for DevOps-style tasks Chaos testing and recovery inside GKE microservices. The importance of making developer UX smooth (dashboard, one-click deploys)

What's next for GreenOps AI Agent

Real-time integration with KubeCost for finer-grained cost analysis. Extending to multi-cluster sustainability reports. Add Self healing agent to take forward Gemini's recommendation AI-driven recommendations (autoscaling, workload scheduling to green regions).

Built With

• gemini-1.5-flash
• python
• streamlit
• vertexai