Insights
Field notes from building platforms
Technical articles on Kubernetes, DevOps, CI/CD, and cloud native engineering.
Building an Enterprise Platform on Bare Metal — Part 5: Running an AI Agent Platform on Kubernetes
The most unexpected part of our bare-metal platform: 18 AI agents, an LLM gateway, and full LLM observability — all on the same three NUCs. Here is how LiteLLM, Langfuse, and a DGX Spark fit together.
Building an Enterprise Platform on Bare Metal — Part 4: Observability with the LGTM Stack
You can’t manage what you can’t see. Here is how we run the full LGTM stack on the same three NUCs as the workloads it observes — multi-tenant, S3-backed on Ceph, with Grafana Alloy as the only agent on every node.
Building an Enterprise Platform on Bare Metal — Part 3: One Repo to Rule 84 Applications
The hardest part of managing 84 applications isn’t deploying them — it’s stopping them from turning into a dumpster fire of configuration drift. Here is the repo layout, the real Argo CD manifests, and the incidents that shaped the patterns.
Building an Enterprise Platform on Bare Metal — Part 2: Bootstrapping with Tinkerbell & Cluster API
How do you provision bare-metal Kubernetes without touching a USB stick? With Tinkerbell for PXE-based OS install and Cluster API turning the whole thing into Kubernetes resources you git commit.
Building an Enterprise Platform on Bare Metal — Part 1: Why We Did It
Everyone told us to use the cloud. We chose three Intel NUCs instead — not to be contrarian, but because the economics made it obvious. Here is the napkin math, the workload, and the hardware.
The Dynamic Duo of Kubernetes: Argo CD + Cluster API
Argo CD knows how to deploy applications. Cluster API knows how to build clusters. Neither knows the other exists. Here is the controller that teaches them — and what the resulting pipeline looks like in practice.