Skip to content

Full-stack · AI agents · Self-hosted

Hi, I'm Roman Pikl.

I build softwarethat actually ships.

I design, build and run complete applications: web and mobile frontends, APIs, databases and the servers they live on. AI coding agents like Claude Code and Codex write much of the code. I set the direction, review what they produce and answer for the result.

agent-run · feature/shift-export
  1. specexport monthly shifts as PDF
  2. plan3 tasks → api · web · tests
  3. agent-1api: /exports/shifts.pdf
  4. agent-2web: download button + states
  5. agent-3tests: edge cases, holidays
  6. verifytypecheck · lint · test
  7. reviewdiff read by a human
  8. deploydocker compose up -d
Illustrative workflow, not a live log

01Selected work

Projects

Most of my repositories are private: client and internal systems. These case studies describe what was built and how, without exposing code or client data. Status labels show what is actually verified.

02Websites

Websites & landing pages

Live sites for clients and for my own products: designed, built and deployed by me, running on my own servers.

03Behind the Build

How software gets built here

Agents are fast, but they are not reliable on their own. The process below is what turns their output into software I am willing to put my name on.

  1. 01

    Requirements

    Talk to the people who will use it. Write down the real workflow, edge cases and what “done” means before any code exists.

    spec.md

  2. 02

    Architecture

    Pick the data model, boundaries and stack. Decide what must be deterministic and tested, and what can iterate.

    schema + ADR

  3. 03

    Agent orchestration

    Split the work into independent tasks and run Claude Code / Codex sessions on them in parallel: separate branches or worktrees, a clear brief each.

    parallel runs

  4. 04

    Review & tests

    Read every diff. Typecheck, lint, unit and integration tests, plus a second agent pass for code review. Anything I cannot explain does not merge.

    CI gate

  5. 05

    Deploy & operate

    Docker images behind Caddy on my own Linux servers, with migrations, backups and logs. Then I watch real usage and iterate.

    docker compose up

Rules I keep

  • The agent writes the code, I own the result.
  • No merge without a test or a reason why not.
  • Small tasks, clear briefs, many parallel runs.
  • Production data never goes into a prompt.

04Tech Stack

Tools I use in real projects

Grouped by layer. Listed only if used in at least one of my own projects.

Web & mobile08

  • TypeScript
  • React
  • Next.js
  • Tailwind CSS
  • PWA
  • Flutter / Dart
  • Kotlin (Android)
  • C# / .NET desktop

Backend & APIs08

  • Node.js
  • NestJS
  • Fastify
  • Python
  • FastAPI
  • REST / OpenAPI
  • SSE
  • BullMQ

Data06

  • PostgreSQL
  • Supabase (RLS)
  • Drizzle
  • SQLAlchemy + Alembic
  • Redis
  • SQLite

Infrastructure06

  • Linux VPS
  • Docker / Compose
  • Caddy
  • WireGuard
  • systemd
  • GitHub Actions

AI & automation07

  • Claude Code
  • OpenAI Codex
  • Claude API (OpenRouter)
  • Ollama (local models)
  • MCP
  • OR-Tools CP-SAT
  • n8n

Hardware & integrations04

  • ESP32 / Arduino
  • QR & NFC check-in
  • GPS fleet units (Wialon)
  • Mapbox

05About

From wiring cabinets to shipping software

I started in electrical engineering: security systems, access control and building automation. That work taught me to think in whole systems, where hardware, software and people have to work together and failures cost real money.

Over time I moved to building the software myself. Today I develop web and mobile applications full-time, and I run my own servers: Linux, Docker, Caddy, PostgreSQL, backups and deploy scripts.

AI coding agents changed how I work, not what I am responsible for. I break problems down, write the specification, let agents implement it in parallel, and then review, test and fix until it really works. My long-term goal is infrastructure where agents can take a larger project from brief to a tested deployment.

Based in
Czech Republic
Background
Electrical & automation systems
Focus
Full-stack · agents · infra
Brand
Apex Piksel (commercial)

06Experimental

Ask PiklDev

A small assistant that answers questions about my projects and experience. This version is a demo: it matches your question against a verified, hand-written knowledge base. No LLM, no tracking, nothing is sent anywhere.

ask.pikldev

Try asking

07Contact

Let's build something that ships.

I'm open to collaboration alongside my current work, flexibly as a contractor. Write to me by e-mail; my public work is on GitHub.