About

Engineering,
end to end.

I build software end to end. In practice that means I am the person who designs the schema, writes the server, writes the client, does the native integration, sets up the deployment, and then owns whatever breaks at 2am - because on most of what I have shipped, there was nobody else to hand it to.

Portrait of Raffael Choi
Raffael ChoiVienna, Austria

Why that shape

Owning every layer of a system is the fastest way I know to learn where the real costs are: the encryption scheme that is correct but unusable, the abstraction that saves ten lines and hides a race, the install step that quietly loses half the people who were going to try it.

Why the range

A game client and a voice agent look nothing alike until you are inside them: both are real-time systems where something arrives on a socket, has to be understood before the next frame or the next word, and where being late is indistinguishable from being wrong. Working across C++, Rust, TypeScript and Python means I pick the language the problem wants rather than the one I already know.

Where the models fit

Machine learning is part of the work and not the point of it. I care about what happens when a model has to answer inside a latency budget, on hardware somebody else owns, with a person on the other end who notices immediately when it is wrong. Those are systems problems before they are modelling ones.

Who I work as

Until October 2026 my projects ran under Phonity Labs e.U. in Vienna. They continue as private, non-commercial projects under the Frostforge label. Not all of it has been solo: at IT-Technik Klimt I led AI automation projects and the cross-functional teams that put those tools in front of people who had not asked for them, and at RISE I worked on configuration management and deployment across distributed infrastructure.

Read the dates carefully

My job titles and my engineering do not share a start date. I spent years in operations roles before software was what it said on the contract, and through four of those years I was also lead PvE developer on Turtle WoW - writing encounters and rebuilding raid content the server core had never implemented correctly - on a server that went from thirty concurrent players to a peak above two hundred thousand. The title caught up in 2023. The work did not start there, and the case studies on this site are where you can check that.

What next

A team doing this at a scale where the interesting problems are not ones I can hold in my head.

Experience

Where I have worked

January 2025 - October 2026

Founder & Software Engineer

Phonity Labs e.U. · Vienna, Austria

  • Built an AI voice agent for privacy-sensitive settings such as medical practices - answering calls, gathering information and booking appointments through webhooks into existing workflows.
  • Owned the telephony layer: PBX and SIP trunk integration, and the speech, language-model and synthesis pipeline behind it, running on local hardware; no cloud version exists.
  • Shipped products end to end under the Frostforge label - Frostphone and Frostify built from nothing, Frostcord taken forward from an MIT fork - owning architecture, implementation, release engineering and the legal surface around them. They continue as private, non-commercial projects.
June 2025 - May 2026

Software Developer

RISE - Research Industrial Systems Engineering · Vienna, Austria (remote)

  • Built internal tooling, automation scripts and integration logic in Ruby and Bash for engineering and research projects.
  • Maintained and extended Puppet configuration management across distributed infrastructure, and coordinated deployments across server pools.
  • Detected and resolved environment drift, improved CI/CD processes and made system behaviour predictable enough to reason about.
December 2023 - May 2025

IT Operations & AI Project Lead

IT-Technik Klimt · Offenburg, Germany (hybrid)

  • Led cross-functional teams designing and deploying AI tooling into working business environments - the part where a model stops being a demo and starts having users who did not ask for it.
  • Ran end-to-end AI project delivery for clients - scoping, planning, implementation and post-deployment support - so the tools kept working after the launch meeting.
  • Ran critical server and network infrastructure with proactive monitoring, incident management, backup strategy and compliance measures against a live threat model.
2019 - 2023

Lead PvE Developer

Turtle WoW · Remote

  • Brought the account and character databases of a discontinued server to Turtle WoW at the end of 2019 - turning down paid offers from competing projects - so that its players kept the characters, guilds and years they had built instead of starting over. Concurrent population went from roughly 30 to over a thousand and compounded from there; the server peaked above 200,000 concurrent players across its EU and CN realms.
  • Rebuilt the vanilla raid encounters, which the MaNGOS-lineage core had never implemented correctly - phase logic, timers, threat behaviour, add handling and ability rotations - then tuned them to match the original fights.
  • Owned custom PvE content - dungeons, raids and quests scripted against the server core and its content database. Shipped Hateforge Quarry, a five-player dungeon with ten custom quests, in patch 1.16.1.
  • Maintained the bridge to upstream VMaNGOS for four years, keeping a heavily customised fork structured so upstream fixes could still be cherry-picked instead of being lost to conflict.
The one-page resume

Stack

What I work with

Languages

  • TypeScript
  • Rust
  • Python
  • C++
  • PHP
  • Kotlin
  • Ruby
  • Bash
  • SQL
  • Lua

Game engines & systems

  • Unreal Engine 5
  • UMG
  • C++ gameplay programming
  • Binary wire protocols
  • Entity streaming
  • Server emulators (VMaNGOS)
  • CMake

Backend & real-time

  • Node.js
  • Bun
  • WebRTC
  • LiveKit
  • SIP telephony
  • Socket.IO
  • Laravel
  • Drizzle ORM
  • PostgreSQL
  • MySQL
  • SQLite
  • Redis

Frontend

  • React
  • Vue
  • Next.js
  • Astro
  • Tailwind CSS

Native & mobile

  • Tauri 2
  • Electron
  • Android (Kotlin)
  • MediaSession & foreground services
  • Widevine EME
  • Raspberry Pi

AI & machine learning

  • Local LLM inference (Ollama)
  • Speech recognition (Whisper)
  • Neural text to speech
  • Tool & function calling
  • PyTorch / TorchScript
  • VRAM & latency budgeting
  • Prompt and role design
  • n8n workflow automation

I build and operate systems around models: local inference, real-time pipelines, latency and VRAM budgets, and what happens when the model is wrong. I have not trained foundation models.

Infrastructure & security

  • Docker & Compose
  • Proxmox / KVM
  • Puppet
  • Linux (Ubuntu, Debian)
  • CI/CD
  • Cloudflare Pages & Workers
  • Caddy / reverse proxies
  • End-to-end encryption
  • ed25519 signing
  • SRP6
  • TOTP two-factor
  • Content Security Policy
  • systemd
Get in touchSee the work