Folio I · Platform & Site Reliability Engineering

Works in production.

I build the Kubernetes platforms, pipelines and observability that let engineers deploy without fear, and I run this page the same way.

See the results ↓ Curriculum vitae

Folio II

About

Most outages are not exotic. They come from slow, frightening deploys, from alerts nobody trusts, and from systems only one person understands. My work is removing those conditions.

I like boring reliability: deploys that are routine, dashboards that answer real questions, and runbooks written for whoever is paged at three in the morning. Everything I build is defined in code and reviewed like code, this page included.

Instruments of the trade: Kubernetes, Terraform, GitOps, GitHub Actions, Prometheus, Grafana, SLOs, incident response.

Folio III

Results

6min

deploy time, formerly 45 min

A pipeline people use

Problem: 30 services, a manual checklist, weekly batches.
Method: Shared CI template, cached builds, canary with auto-rollback.
Result: Several deploys a day instead of one a week.

−60%

on-call pages, ≈23 to ≈9 a week

Alerts on symptoms

Problem: Hundreds of threshold alerts, mostly ignored.
Method: SLOs per user journey; page only on error-budget burn.
Result: Fewer pages, no missed incidents.

−38%

compute spend

Right-sized clusters

Problem: Sized for peak, all year round.
Method: Measured requests vs limits, autoscaling, spot for batch work.
Result: Same latency, much smaller bill.

hand-built resources, formerly ≈400

Everything as code

Problem: Cloud resources made by hand, understood by few.
Method: Imported into Terraform module by module; plans reviewed on every PR.
Result: Nightly drift detection, nothing hand-made left.

Folio IV

Status

99.98%

availability over the last 30 days, against an objective of 99.9 %

Measurements
Window	Uptime	p95 latency
24 hours	100.00 %	142 ms
7 days	100.00 %	151 ms
30 days	99.98 %	158 ms

All systems operational. Read every five minutes by an independent observer outside the machine; full record.

Folio V

Stack

Scroll to take the machine apart; scroll back to rebuild it.

Parts of the machine
Part	Office	€/mo
Cloudflare	DNS, TLS, WAF, cache, tunnel	0.00
Hetzner CAX11	2 vCPU ARM server, no inbound ports	4.50
cloudflared	opens the tunnel from inside	0.00
nginx	serves this page, read-only	0.00
systemd timer	fetches each new image within 60 s	0.00
GitHub Actions	builds, checks, publishes, verifies	0.00
Terraform	describes every part above	0.00
Upptime	the independent observer	0.00
Total		€4.50

Next folio: the cabinet becomes three, joined into a Kubernetes cluster with GitOps, metrics and published objectives.