Hi, I'm Antonie 👋

I am a software engineer and engineering lead with more than 12 years of experience. I build web products, backend services, cloud platforms, and applied AI systems.

I have shipped software for browsers, mobile apps, smart TVs, game consoles, and cloud services. I also ask, “What happens when this fails?” earlier than most people would like.

const antonie = {
  builds: ["platforms", "web products", "AI systems"],
  likes: ["clear contracts", "small deploys", "useful logs"],
  currentlyExploring: ["reliable AI agents", "Go", "local LLMs", "civic tech"],
  recurringQuestion: "Can we make this simpler?",
};

🛠️ What I build

  • APIs and backend services that stay clear after the first release
  • React and Next.js products with reliable backend and platform support
  • AI workflows with structured outputs, evaluations, and traceable results
  • Data pipelines that recover from failure without repeating costly work
  • Internal tools that remove routine work instead of adding another dashboard

Core tools: TypeScript, Node.js, C#/.NET, React, SQL, MongoDB, AWS, Azure, Docker, and Kubernetes.

Also using: Python, Go, OpenAI and Gemini APIs, RAG, local models, n8n, and far too many JSON schemas.

🧪 Current projects and side quests

  • LedgerFlow — A document intelligence and invoice automation copilot for small ERP and accounting teams.
  • LocalScout — An offline-first React Native app for field research and local directories.
  • Patokaz — A civic technology and open-data platform that helps people find services they can trust.
  • Budno — A research project that uses citizen reports, satellite data, and computer vision to study local environmental problems.
  • Local AI and workflow automation — Local models and n8n workflows that run on small Apple Silicon computers.
  • Data-driven web projects — Searchable web applications that turn messy public data into useful information.

Patokaz and Budno are still evolving. I share code when it becomes useful and safe to run in public.

🧠 What production incidents taught me

Open my production incident checklist
  • Time zones need tests. Daylight-saving changes need more tests.
  • A retry without idempotency can turn one failure into several wrong results.
  • “Works in Chrome” does not mean “works on a 2018 smart TV.”
  • If you cannot trace an AI answer to its inputs, you have a demo, not a system.
  • Teams usually add the most useful log one incident too late.

☕ Say hello

Want to talk about system design, applied AI, streaming platforms, developer tools, open data, or civic software for North Macedonia? Send me a note.