Senior DevOps Engineer · MLOps / AI enthusiast · NixOS tinkerer

I build reproducible infrastructure, GitOps platforms, automation workflows, MLOps tooling, and local-first developer environments.

I am a Senior DevOps Engineer from the Netherlands. My work lives around infrastructure automation, platform engineering, observability, and AI/MLOps systems.

I like infrastructure that is boring in production and interesting in design: declarative, versioned, tested, observable, and recoverable. I prefer systems where the desired state is reviewable in Git and where recovery is part of the architecture, not an afterthought.

No clicky-click infrastructure. If it matters, it should be declarative, versioned, tested, observable, and recoverable.

I usually work where infrastructure, developer workflows and AI systems meet: platforms that need to be automated, observable and recoverable, but still nice to work on every day.

Platform engineering: cloud, datacenter and self-hosted platforms built around IaC, GitOps, orchestration, CI/CD, secrets, policy checks and monitoring.

MLOps / AI systems: local models, hosted APIs, routing layers, retrieval, agent tooling, traces, evaluation loops and observability for AI workflows.

Reproducible systems: NixOS, Home Manager, homelab automation, backup strategy, rootfs-based containers, recovery flows and local-first tooling.

Graft is my current main open-source project.

Graft is a TOML-driven workflow for building Podman Quadlet containers from the Nix store. The goal is to keep container intent small and readable while letting Nix materialise the rootfs and Quadlet output.

Why it exists: I want development and service containers that are declared like infrastructure. Packages come from Nix, runtime output is generated, and the user does not hand-write Quadlet boilerplate or install tools ad-hoc inside containers.

Current alpha: v0.1.0-alpha.1

Current scope: TOML to JSON resolver, NixOS system containers, Home Manager user containers, rootfs-store materialisation, graft-pause, CI and docs.

This is the compact version of the stack I actively work with or experiment with. The long version changes often; the themes do not.

Cloud and platforms Azure · AWS · IBM Cloud · datacenter · Proxmox · TrueNAS · NixOS · macOS

Containers and orchestration Kubernetes · Talos · Docker · Swarm · Podman · Quadlet · Nomad · Helm · GitOps

IaC and automation OpenTofu · Terraform · Bicep · ARM · Ansible · Python · PowerShell · Bash · TypeScript

CI/CD and quality GitHub Actions · Azure DevOps · Jenkins · Forgejo · unit tests · linters · policy checks

Observability Grafana · Prometheus · Loki · Datadog · Splunk · OpenTelemetry · Azure Monitor · Langfuse · Phoenix

Secrets and security Infisical · Azure Key Vault · SOPS · age · Vault-style workflows · least privilege · GitOps-safe secrets

MLOps and AI OpenAI · Claude · Azure AI · IBM Cloud AI · Hugging Face · Ollama · llama.cpp · vLLM · LiteLLM · Bifrost · LangGraph

Data and retrieval PostgreSQL · SQL Server · MySQL · MongoDB · Redis · Qdrant · vector search · BM25 · hybrid retrieval

I am especially interested in the practical side of AI systems: how models, retrieval, prompts, traces, evaluations, routing, local inference and production infrastructure fit together.

Local-first AI: Ollama, llama.cpp, vLLM and local model stacks for experimentation, privacy, latency and control.

AI platform tooling: LiteLLM, Bifrost, LangGraph, SDKs, hosted LLM APIs and routing layers that make AI workflows operable.

Observability: Langfuse, Phoenix, OpenTelemetry-style thinking, traces, evaluation loops and visibility into what AI systems are actually doing.

infrastructure automation
  + developer tooling
  + local-first systems
  + AI-assisted workflows

I run a homelab that is managed like real infrastructure, not like a pile of manually configured machines. It is where I test platform ideas, GitOps flows, backup strategies, HA patterns and AI infrastructure before they become muscle memory.

Platform shape: Proxmox, Talos, Kubernetes, Nomad, devcontainers, GitOps, Ansible and OpenTofu. The goal is to keep services reproducible and rebuildable instead of precious snowflakes.

Reliability shape: TrueNAS mirrors, Proxmox Backup Server, 3-2-1 backups, 10Gbit backplane, Technitium DNS HA, monitoring, alerting and recovery-oriented design.

Services and patterns include Traefik, Authentik / Keycloak-style identity, Nextcloud, DDNS, firewalling, Prometheus, Grafana, Loki, Kuma and more.

My daily setup is terminal-first and declarative where possible.

NixOS · Home Manager · Niri · macOS · Kitty · tmux · Neovim · Git

The goal is a machine that can be rebuilt, understood, versioned and tuned without turning the workstation into undocumented state.

Declarative first: if it matters, it belongs in config. I prefer reviewable desired state over manual changes and hidden drift.

Observable by default: logs, metrics, traces and health signals should be part of the design, not a panic-driven retrofit.

Recovery matters: backups, rebuilds and rollback paths are as important as deployment pipelines. A system is not done until it can recover.

Small tools, clear boundaries: I like tools that do one job clearly and can be composed into bigger workflows without hiding state.

NixOS as infrastructure · rootfs-based containers · Podman Quadlet · reproducible developer environments · secure GitOps · local LLM stacks · MLOps observability · AI agents for developer workflows · homelab-to-production patterns

The best place to reach me for now is GitHub.

I am gradually publishing reusable parts of my private infrastructure and tooling as open-source projects.