← CodePopThe founder's public work
Engineering
you can inspect.
Public projects by Aleksandar Popovic, CodePop's founder. Each case connects a concrete engineering problem with the implementation, its current boundaries, and evidence you can inspect.
From a problem to a working boundary
These are open-source engineering case studies. The linked repositories and local verification captures establish the scope of each example. They do not represent unnamed client engagements or production adoption figures.
01 / Container orchestration
orch
The problem
A small Docker fleet still needs scheduling, reconciliation, and a clear deployment history. Running containers by hand leaves those responsibilities scattered across scripts and individual machines.
The engineering decision
A CLI talks to the server API. The control plane schedules work and reconciles desired state; node agents poll for assignments and operate Docker. Logs, events, and rollout status expose what each component actually observed.
What the evidence establishes
The repository provides a local development stack, operator commands, and a concrete separation between control-plane decisions and node execution. The README identifies restart-safe server state wiring as further work; a PostgreSQL store alone is not evidence of a durable runtime.
Documentation checked October 12, 2026 against source cbfc94a ↗.
Explore the implementation and demo ↗
02 / Kubernetes CI/CD
cloudivision
The problem
Build execution and environment deployment need distinct ownership. A CI runner with broad cluster permissions makes that boundary difficult to review and recover.
The engineering decision
Source events create build resources in Kubernetes. Controllers reconcile isolated runner Jobs, which produce image evidence. A deployment repository records the release intent, and Argo CD or Flux reconciles the environment.
What the evidence establishes
The project documents a kind quickstart, an operations UI, and a credential-free example. Its published alpha has explicit readiness gates; development authentication settings require review before evaluating a production installation.
Documentation checked October 12, 2026 against source 7eb7739 ↗.
Explore the implementation and demo ↗
03 / Platform infrastructure
ClusterForge
The problem
Repeated platform setup needs reusable conventions while preserving the infrastructure files an operator must inspect, change, and review.
The engineering decision
A Go CLI generates projects and environments from focused foundation, orchestrator, platform, and workload modules. Generated Terraform remains visible. Planning, policy checks, and operational evidence sit alongside explicit apply workflows.
What the evidence establishes
The repository includes modules, examples, installation guidance, and a local Kubernetes workflow. Platform coverage has distinct maturity levels; each real cloud environment still needs its own network, IAM, deployment, and recovery evidence.
Documentation checked October 12, 2026 against source 7996921 ↗.
Explore the implementation and demo ↗
04 / Docker over SSH
ContainerDesk
The problem
Inspecting several remote Docker hosts means switching terminals and remembering each connection context. Management actions also need an unmistakable target and permission boundary.
The engineering decision
The desktop app reuses OpenSSH configuration, keys, agents, and host verification. React presents workloads and logs while the Rust backend enforces permissions. Recovered and new sessions begin with inspection access; management and terminals are granted explicitly.
What the evidence establishes
The project publishes Linux and macOS preview packages and records native verification evidence. Containers, logs, and existing Compose groups share one workspace. Images, volumes, and networks remain read-only within the documented preview scope.
Documentation checked October 12, 2026 against source 499545b ↗.
Explore the implementation and demo ↗
Bring your system's constraints
A useful project brief identifies the workflow, current system, failure mode, and outcome you need. For a delivery system, that might mean a release that can be traced and recovered. For an operator tool, it might mean knowing exactly which host an action will affect.
Discuss your engineering problem ↗