Azure DevOps
Engineer
Patrick Koorevaar is a senior Azure DevOps Engineer focused on cloud-native infrastructure, DevOps automation, and container orchestration. At Q-Vend B.V. he leads end-to-end Azure infrastructure, automation, and scalable deployments across AKS, ACA, CI/CD, and production platforms. In his side projects, he has built a delivery pipeline in which AI agents do the routine work, with the guardrails in code and the final review his own. Currently studying for the CKA.
Before publishing full projects on GitHub, I spent years helping others on Stack Overflow: 59 questions answered, 1.4k reputation, mostly C#, Git, and Azure/Kubernetes scenarios.
Every side project starts from "what can the free tiers cover" — Cloudflare Workers AI, Cosmos DB free tier, Container Apps scaled to zero. VS Code for frontend, VS Pro for backends, same split at work and on side projects.
Full toolset and technologies in the Skills section below.
Applied AI, running live
Gated RAG assistant on Cloudflare Workers
Answers visitors' questions from a landing page's own content. The contact offer is decided by code, not the model, and only after a few genuine questions.
Instagram reply agent
Claude Haiku drafts replies to Instagram Reel comments in my tone. Nothing is posted until I approve it on a page behind Cloudflare Access.
"Ask about Patrick" chat assistant
The floating chat button on this page — a retrieval-augmented assistant that answers questions about my background from a knowledge base built entirely by Claude Code.
Blog comment moderation
Every comment on the blog is scored by an LLM via a Cosmos DB change-feed-triggered Azure Function, calling Cloudflare Workers AI cross-cloud.
AKS/ACA capacity advisor
Turns cluster telemetry into plain-language capacity recommendations using Workers AI.
AI text analysis
Sentiment and language detection served live over a Cloudflare Workers AI endpoint.
Multi-agent hangman
Two Workers AI agents compete live; game state streamed over SignalR on Azure Container Apps. Join in as a third human player.
CV request generator
A button on my contact page asks an LLM to write a personalized message requesting my CV — an AI gimmick that's actually live.
Other side projects
Infrastructure & DevOps
Production AKS at scale
30+ microservices per namespace on production AKS clusters, including two full blue/green cluster migrations (2022, 2025).
CI/CD pipelines
Azure DevOps YAML pipelines, automated test with a manual production gate, plus blue/green and canary rollouts via Traefik and feature flags.
Infrastructure as code
Terraform and Bicep, deployed via GitHub Actions with OIDC and least-privilege roles kept out of the CI pipeline by design.
K8s cluster automation
Fully automated two-node Kubernetes cluster in an Azure sandbox, provisioned with Bash and Bicep, standing up end-to-end in under 7 minutes with Calico CNI over VXLAN.
CKA practice lab
A real Linux shell relayed live to a cloudshell-style browser terminal over SignalR, shimmed kubectl/kubeadm, gated by Cloudflare Zero Trust.
Skills
Certifications
My work style
Building with AI, not just prompting it
I build my automation in a back and forth with Claude Code. It proposes the design, writes the code, tests and docs, and asks me numbered questions with a recommendation when a decision is mine. I create what only I can (Cloudflare tokens, GitHub environments and secrets), review every PR, and push back when something doesn't fit. In one day of that loop we built an end to end, AI-operated deployment for client sites: I give Claude a site name and a request, and after my review and merge the pipeline puts it live.
Agentic backlog management
Ideas and the feature backlog live on a private GitHub Projects board. Claude Code files and updates items directly via the GitHub GraphQL API during planning conversations, capturing decisions and open questions as they're made.
Blog content pipeline
Posts are written as Markdown, stored in Cosmos DB, served through an ACA API. I ask Claude to insert or update articles directly; deleting a post is deliberately left to me only. A Cloudflare Worker converts Markdown to HTML on the way out.
Building this website
Started out built with Aider paired with a local Qwen2.5 model running via IPEX Ollama. Ongoing iteration — new sections, project cards, fixes — now runs through Claude Code instead.
Everyday tool stack
VS Code for frontend work, Visual Studio Pro for backends — the same split at work and on side projects.
Cost-conscious architecture
Every side project starts from "what can the free tiers cover" — Cloudflare Workers/DNS/Web Analytics, Cosmos DB free tier, Container Apps scaled to zero, Entra ID plus Cloudflare Zero Trust for auth. Workers AI over Azure AI Foundry for the same reason. Anything bigger runs briefly on the Azure credit from my VS Pro subscription, then gets torn down.
Backlog
Nothing on the backlog right now — check back later.