Work
Work
Two tracks that have been running in parallel for years. Pick one.
Unreal Engine: game code, not prototypes.
DualSense Adaptive Triggers
UE5 plugin · Gumroad
Unofficial Unreal Engine 5 plugin bringing the DualSense controller's adaptive triggers and haptics into the engine, driveable straight from Blueprint.
02The DLSS fix, before NVIDIA
NVIDIA Developer Forums
Unreal Engine 5.5 wouldn't compile against DLSS 3.7.20. I posted the fix on the NVIDIA developer forums before the official patch landed.
How I work
C++ and Blueprint
I write C++ and expose it to Blueprint so designers can work without asking my permission. Extendable components, built to be reused in games other than the one they were written for.
Profiling and performance
Unreal's profiler used properly: finding the real bottleneck instead of optimizing on instinct, and holding the frame rate steady once the scene fills up.
From design to build
Taking a concept to something playable without losing the original idea along the way. This is where creative vision and technical constraints have to reach an agreement.
Working in a team
Short iterations, project state always legible, fewer meetings. A team that knows where it stands moves faster than a better team that doesn't.
Helping other developers
On Fiverr I've unblocked other people's projects: bugs that wouldn't reproduce, pipelines that needed rewriting, performance to claw back. The fastest way I know to test what you actually know.
Web, utilities, terminal things. What I build when the problem is mine.
Unreal-TUI
Rust · ratatui · crossterm
Cross-platform terminal UI in Rust for reading Unreal Engine logs in real time: live tailing, category filters with clickable tokens, timestamp toggle, word wrap, and automatic project and build discovery from running editor processes.
02X11Vm
TypeScript · Tauri · Docker · Xpra
Tauri desktop app that manages and runs Dockerized X11 environments. Docker, Xpra and Tauri together, so containerized applications run as rootless native windows on Windows, macOS and Linux — the XQuartz idea, taken off the Mac.
03Cited by Claude
GitHub Gist
A solution of mine surfaced in a public GitHub Gist thread, cited in Claude's answer. Small win, kept for the record.
The method
Five phases. I don't skip them, however hard someone pushes.
Prototype and MVP
The smallest version that proves the idea stands up. Its job is to reveal what isn't needed: half the initial requirements don't survive the first working prototype.
Architecture design
Service boundaries, the shape of the data, where splitting helps and where splitting just costs. The uncomfortable decisions get made here, while changing them is still free.
Environments and release pipeline
Reproducible environments, GitLab CI pipelines, delivery onto Kubernetes clusters. A release should be boring: if deploying is an event, something upstream is broken.
Post-release monitoring
Sentry for errors, metrics on what actually matters. The moment you find out about a problem has to be before the client's phone call, not after.
Resource optimization
Distributed compute where the load justifies it, async queues for everything that doesn't need to answer immediately. Most performance problems are work done in the wrong place, not work done slowly.