Bucharest, Romania

Cristian Glavan

Engineering Manager, Infrastructure & Platform Engineering at Zitec.

AI

Beyond infrastructure, I'm deeply drawn to AI — not just using it, but understanding how models are trained, fine-tuned, and deployed.

aitraining.glavan.tech
···
All-time downloads
Hugging Face · Open Weights

magos-k8s-0.6b

A Kubernetes debugging assistant that runs locally.

A 0.6B-parameter model fine-tuned from Qwen3-0.6B on Kubernetes documentation, the full Kubernetes API reference, the kubectl command reference, and Prometheus alert runbooks. It builds kubectl commands, generates valid YAML manifests, and diagnoses errors pasted straight from logs and alerts — small enough to run on your laptop and be embedded in autonomous DevOps agents.

0.6B params Qwen3-0.6B base GGUF · q4 / q8 / f16 Apache 2.0
KubeHealth-Bench · 54 real incidents Scorecard →
Names the right root cause87%
Claude Haiku 5.5 99% · Gemini 3.8 Flash 89%
Names the decisive detail89%
Claude Haiku 5.5 89% · Gemini 3.8 Flash 93%
Gives the right fix87%
Claude Haiku 5.5 93% · Gemini 3.8 Flash 89%
Finds a cause in another object78%
Claude Haiku 5.5 44% · Gemini 3.8 Flash 62%
magos v19 + links vs hosted models, mean of 3 seeds.
View on Hugging Face
Open Source · Kubernetes Triage

KubeHealth

See what's broken, not a wall of events.

One binary you point at your kubeconfig. It groups thousands of warning events into one row per broken workload, ranks them by how new they are, and can ask magos — bundled and running offline on your laptop — for a root cause using read-only tools. Nothing is installed in the cluster. Prefer a frontier model? One switch sends the analysis through Claude Code instead.

Single binary Read-only Offline AI · magos bundled Claude Code integration
clglavan.github.io/kubehealth

About

I lead the DevOps and CloudOps teams at Zitec. We help companies design, build, and run their cloud infrastructure — migrations, infrastructure as code, CI/CD pipelines, containerizing applications onto Kubernetes, and day-to-day operations across AWS, GCP, Azure, and on-prem.

We work as a team extension for engineering orgs that already have DevOps in-house, or as the full ops team for companies that don't. Engagements range from short audits and proofs of concept to long-running managed services.

Things I personally like

Contact