Overview
Tech stack
Every technology in the platform, and the reasoning behind choosing it. Each pick is deliberate: chosen for what it proves as much as what it does.
The stack leans heavily on CNCF Graduated and Sandbox projects. That's intentional: it keeps the platform aligned with where the ecosystem is going, and every choice doubles as a signal to anyone reading the code.
Choices and rationale
| Layer | Technology | Why |
|---|---|---|
| Cloud provider | Civo | Cheap, fast K3s clusters and a CNCF member. Keeps the project from being just-another-EKS-demo. |
| Infrastructure as Code | Terraform / OpenTofu | Industry standard, reusable modules, the foundation the whole platform is provisioned from. |
| Kubernetes distribution | K3s (via Civo) | Lightweight and production-ready, with a good story to tell in write-ups. |
| GitOps | Argo CD | CNCF Graduated, the most adopted GitOps tool, and a strong resume signal. Manages itself and everything else. |
| Service mesh | Istio | CNCF Graduated. Adds mTLS, traffic management, and observability: a deliberate Staff-level complexity signal. |
| Monitoring | Prometheus | CNCF Graduated, the de facto Kubernetes monitoring standard. |
| Dashboards & alerting | Grafana | Industry standard, pairs directly with Prometheus. |
| Demo workload | Go (crashdummy) | A chaos-enabled mock backend. It brings Go (the language of the K8s ecosystem) to the portfolio. |
| AI SDK | Anthropic SDK (Python) | Direct, no framework overhead. The right place to start the agent work. |
| AI on K8s framework | kagent | CNCF Sandbox, Kubernetes-native agent framework. Demonstrates awareness of the emerging ecosystem. |
| Secure execution | Agent Sandbox / gVisor | Sandboxed code execution for AI agents, with no host kernel access. |
| Agent protocol | MCP | The Model Context Protocol: the standard for agent-to-tool communication. |