About
My name is
Andre Semek.
I live in Berlin and work as a Cloud Engineer at Vattenfall. 15 years of infrastructure — from Windows servers through Linux to Azure and Kubernetes.
At Vattenfall.
I'm part of the ACAE team — Azure Container App Environment. We operate a central container platform for more than ten internal business units. My responsibility is the template repository: Bicep modules, Azure DevOps pipelines, DTAP model (Dev/Test/Acc/Prod), Blue-Green deployments with traffic splitting, Key Vault integration.
Before going Azure-native: classic Windows and Linux server administration, virtualisation, containers. The line from then to now is straight: maintainable setups, small diffs, clear modules, documented standards.
Personal.
Two kids, football, music.
ADHD — means I work better with clear structures, short cycles, and a lot of documentation. That shapes how I run projects: small steps over big leaps, early showing over big reveals at the end.
Small steps over big leaps, early showing over a big reveal at the end.
What I'm working on right now.
Azure & Container
Container Apps, AKS, ACR, App Services, Key Vault
Infrastructure-as-Code
Bicep, Terraform, Ansible — modules over monoliths
CI/CD & Pipelines
Azure DevOps, GitHub Actions — DTAP, Blue-Green, Rollback
Infrastructure automation
Cloud-hybrid systems, monitoring with Prometheus/Grafana, CLI workflows
AI automation
Several agents running in parallel, cost-aware model choice, MCP servers, custom agent setups
How I work
My approach to AI-assisted work.
I run several AI agents in parallel for my own work, with a fixed division of labor. What follows isn't a sales pitch — it describes how I actually work.
Cost-tier routing
Model selection
Search, renaming, summarizing — a fast, cheap model handles that fine. Architecture or reasoning questions go to the strongest one available. Keeps the budget in check without cutting corners on the hard parts.
Parallel subagents
Orchestration
Research, review and refactoring don't have to run one after another if they're independent. An orchestrator splits the work and pulls the results back together.
Adversarial review
Quality assurance
For bigger architecture decisions, I have a second, independent model check the reasoning before I commit — especially when a later correction would be expensive.
Cross-session memory
Continuity
When a session runs long or the context fills up, a handoff mechanism writes out the state and open items. The next session can pick up right there.
The flow
In client projects I work by the same principles: cost-tier routing, independent counter-review, small diffs, clean handoffs. Which tools sit behind that — Claude, OpenCode, others — is visible in the projects themselves, for instance Pridovo, context-parachute and lore.