Aleksander Cynarski

Platform & DevOps engineer — I turn processes into code, then let agents run them. Gliwice, PL · DevOps Expert @ PKO Bank Polski · GitLab since 2022

🧭 What I do

I build internal developer platforms in a heavily regulated environment — self-hosted GitLab, OpenShift, ArgoCD, Terraform/OpenTofu — and I keep pushing the same idea everywhere: if it is a process, it can be a repository. Pipelines, forms, policies, agent behaviour, documentation, even the catalog of tools you are allowed to use. Most of my public work orbits three things:

• ⚙️ GitLab CI/CD as a product — reusable process pipelines, job contracts, policy-as-code, local pipeline execution.
• 🤖 Agents with an actual contract — LLM agents whose behaviour comes from versioned packages, not from prompts buried in code.
• 🧑‍💻 Developer experience — if a platform needs a manual, it needs a better UI first.

"You won't lose your job to AI, you'll lose your job to somebody who uses AI."

🗺️ The map

flowchart LR
    ME(("automancer"))
    subgraph CI["CI/CD as a product"]
        GLCI["glci — local GitLab CI, Go"]
        SKILLS["gitlab-skills — APM package"]
        MRGEN["cookiecutter-mr-generator"]
    end
    subgraph AGENTS["Agentic engineering"]
        CODEMOD["agentic-codemod-pipeline"]
        ORWIL["orwil — offline voice assistant"]
        MEMVID["memvid — local RAG"]
    end
    subgraph DX["Developer experience"]
        MONAD["Wintermute — JSON forms to pipelines"]
        ARRAKIS["Arrakis — repo architecture graph"]
        VLTED["vlted — Vault CLI"]
    end
    subgraph LAB["Lab and hardware"]
        ESP["esp32 e-ink + gesture"]
        BOX["moadib, mssql — test containers"]
    end
    subgraph WRITE["Writing"]
        BLOG["cynar.ski"]
        EIAC["eiac.dev"]
    end
    ME --> CI
    ME --> AGENTS
    ME --> DX
    ME --> LAB
    ME --> WRITE

🛠️ Stack

Languages Platform & delivery Security & quality AI / agents

🚀 Selected work

1. glci — GitLab CI, run locally

gitlab.com/paramah/glci · glci parses .gitlab-ci.yml, spins up a mock GitLab API and a containerized runner, and executes jobs locally with full runner fidelity. I contributed first-class Podman support — an engine abstraction ([docker] engine = "podman"), socket and connection resolution, glci doctor checks, DinD behaviour and the docs — merged upstream into GitLab's official repository as !141 (labels: 1st contribution, Community contribution).

2. Wintermute — pipelines behind a form

automancer/monad · A client app that turns GitLab repositories into runnable processes. Forms are generated at runtime from JSON files in the repo (.platform/form/), with dynamic fields fed by GitLab REST/GraphQL or any external API, field dependencies, DTO mappers, automatic GraphQL pagination and cross-repository pipeline triggers. Runtime-swappable configuration, hash routing, deep links, Radix UI, dark mode — Clean Architecture end to end.

3. Arrakis — see the architecture you actually have

automancer/arrakis · Browser-only visualization of GitLab repository architecture: dependency graph, health scoring, security scan and code preview. OAuth Authorization Code + PKCE with no backend, batched GraphQL file fetching with a REST fallback, tree-sitter/acorn parsers, and layer boundaries enforced by dependency-cruiser.

4. agentic-codemod-pipeline — agents with a contract

automancer/tests/agentic-codemod-pipeline · A reference repo for an agent network that modifies code, driven from GitLab CI. Scout → planner → coder → reviewer → scribe, with behaviour supplied by APM packages (.apm/agents, skills, instructions, prompts, context) rather than hardcoded prompts — apm-policy.yml constrains what may be installed, apm.lock.yaml pins it. Deterministic steps (ecosystem detection, version bump, tests, git) stay in code; the model only gets what code cannot do. Runs fully offline for demos and tests.

5. gitlab-ci-pipelines — teaching agents our CI

automancer/process/gitlab-skills · An APM package that teaches AI coding agents how to author, debug and optimize GitLab CI/CD — combining GitLab's own MIT-licensed CI skill with my layered process architecture: a reusable common/ + technology/ pipeline, job contracts composed via !reference, snapshot/release workflow typing, and policy validation with OPA Rego, conftest and shellcheck.

6. orwil — a voice assistant that never phones home

automancer/orwil · Fully offline assistant: Whisper for speech recognition, Ollama for conversation, four interchangeable TTS engines (SpeechT5, MMS, Piper, Coqui with voice cloning), first-class Polish support and Apple Silicon (MPS) optimization. Rewritten in 3.0 around Clean Architecture.

📦 The rest of the workshop

Everything below lives under the automancer group — my public lab for platform building blocks.

✍️ Writing

cynar.ski 🧪

Developer ☣️ DevOps ☣️ Freak — my brain is my enemy, enjoy the ride. My personal tech blog (Polish, Hugo, dark by default). Hands-on notes from the lab:

• Serwer ACME lokalnie w labie — homelab PKI
• Zabawa gopass i procesami — secrets, git, GPG
• Testowanie skryptów powłoki — Bats and shell QA
• Shell w procesach CI/CD — error handling that survives production
• Migracja na Hyprland — desktop as configuration
• IoT, czyli prosta elektronika for fun — ESP32 tinkering

eiac.dev 📐

Everything Is A Code — when everything becomes code. A publication and a catalog about running infrastructure, applications, design and the whole SDLC from code. Six pillars, 100+ curated tools: Infrastructure · Applications · Design · SDLC · Security · Observability. Recent pieces:

• Platform engineering: normalizacja pracy w organizacji
• Pamięć agenta jako fakt — nine providers, one class of mistake
• Architektura czasoprzestrzenna: system jako jeden graf
• Pipeline bez sekretów: SPIFFE zamiast hasła w CI

📊 Snapshot

Let's talk platforms. → cynar.ski · eiac.dev · LinkedIn Everything is a code — including this page.