Kubernetes Delivery Platform — CI/CD, IaC & Observability
A delivery platform that takes a containerised app from git push to a running, monitored Kubernetes deployment. The application (Linkly, a URL shortener with a TypeScript API and PostgreSQL) is deliberately simple; the work is everything around it.
View the code on GitHub- Role
- DevOps engineer
- Context
- Personal project
- Period
- 2026
Screenshots & diagrams
Captured from the running project. Click to view full screen.
The problem
Shipping a web app reliably needs more than code: repeatable infrastructure, automated testing, secure containers, safe rollouts and monitoring.
The solution
Built a production-style platform around a small URL-shortener app: hardened containers, Kustomize overlays, Terraform-provisioned cluster with ingress and monitoring, and a 5-stage GitHub Actions pipeline that deploys to a real Kubernetes cluster.
The impact
A pipeline designed to test, scan and deploy every push to Kubernetes, then restart the API under live traffic and fail if a single request errors — zero downtime proven, not assumed.
What I did
- 01
Wrote the API with health/readiness probes, Prometheus metrics and graceful SIGTERM shutdown, covered by integration tests against PostgreSQL.
- 02
Built multi-stage, non-root Docker images and a Docker Compose setup for local development.
- 03
Wrote Kubernetes manifests with Kustomize (dev / prod): rolling updates, HPA, PodDisruptionBudget, StatefulSet, Ingress.
- 04
Hardened the cluster: restricted Pod Security Standard, read-only filesystems, dropped capabilities, default-deny NetworkPolicies.
- 05
Provisioned the cluster with Terraform (kind + ingress-nginx, metrics-server, kube-prometheus-stack via Helm).
- 06
Added observability as code: ServiceMonitor, alert rules linked to a runbook, Grafana dashboard.
- 07
Built the CI/CD pipeline: tests, manifest validation, Terraform checks, Trivy image scan, GHCR publish, e2e deploy to kind with a zero-downtime check.
Key challenges
Zero downtime, proven
It takes four pieces working together — maxUnavailable 0, readiness probes, a preStop delay and graceful shutdown in the app — so the pipeline tests it under load instead of assuming it.
Probes that don't make outages worse
Liveness never checks the database: if Postgres goes down, pods leave the load balancer via readiness instead of being restarted in a loop.
Results
- One command brings up the full platform locally (Terraform + kind).
- 5-stage pipeline covering tests, validation, security scanning, publishing and real Kubernetes deployment.
- Architecture decisions and an on-call runbook documented in the repository.
Tech stack
Skills demonstrated
Want this kind of work on your team?
Available for a 6-month PFE internship from January 2027.