Live

Portfolio V2

Explore how AI works Coming Soon

JustFoura

Fheo / Onion

Projects

Llama Studio V3

On GitHub

A Windows 11/Linux compatible GUI for llama.cpp and vLLM using tauri v2. Rust based. Efficient display.

Why I made it: To provide myself with a easy-to-use GUI for llama.cpp without having to go into the terminal.

What I learned: Tauri applications. llama.cpp. Local model inference.

Personal

Wplace Leaderbot

On GitHub

A Python-based bot to gather daily, weekly, monthly, or all time information for specific configured zones within Wplace.live and display them either on a daily-based timer or on command into Discord via a Discord bot.

Why I made it: To set up an easy automated display of leaderboard scores in specific zones.

What I learned: Discord bot integration. Python. Public information scraping.

Personal

Faster-Than-Light Patches

On GitHub

A GitHub repo filled with documentation and written patches for llama.cpp to run specific LLM models using SSD offloading

Why I made it: To run Qwen3.8 Flash Next on my computer without needing the required hardware.

What I learned: Expert offloading. Model quantization. LLM Prefill and Decode.

Half-Personal

Genxe Game Engine

In development

A Rust / C# based game engine.

Why I made it: To create easy games using AI-MCP integration.

What I learned: To Be Determined.

Half-Personal

Arrow

In development

An always on AI assistant capable of all sorts of tasks ranging from coding to personal assistance.

Why I made it: To have an open-source personalized local AI-assistant capable of browser use, computer use, coding, assistance, etc.

What I learned: Local AI. Databases. Monorepos.

Half-Personal

Minecraft-MCP

On GitHub

An AI-MCP mod for Minecraft Java 26.3 to give AI the ability to autonomously debug and develop Minecraft mods.

Why I made it: To let AI Agents autonomously debug and develop Minecraft mods.

What I learned: Nothing, existing knowledge.

Personal

3D-Place

In development

A 3D three.js web game for building minecraft-style on a map of Earth.

Why I made it: IDK, because I can.

What I learned: Three.js.

Personal

Xenite

Deprecated

A custom AI coding harness, built into a GUI app and a CLI.

Why I made it: To create games and coding projects easier

What I learned: AI-integration. Electron desktop applications. Coding harnesses.

Personal

Tranquility Code

Deprecated

A full TUI AI coding harness built on top of PI coding agent.

Why I made it: To code internally easier using specific trained LLM's.

What I learned: Rust-based TUI. TUI's. Coding harnesses.

Internal

Questions

Based in
New Zealand
Age
๐Ÿคซ
Job
No

Which AI do I use

Depends. I found GLM-5.3 Flash very good at local model optimisations, while GPT-5.6 Luna or Qwen3.8 27B is good at coding. Interestingly, only Qwen3.8 27B was any good at UI.

Any form of AI / coding education

Nope, either self-taught or I let AI take care of it.

AI subscriptions

Currently GPT Plus and GLM Coding Plan Lite, though that's subject to change depending on the latest AI landscape.

Image generation

I tend to avoid it, but for the occasional specific wallpaper or emoji, I might.

Operating system

Windows 11 mainly, sometimes Linux โ€” it has exclusive support for certain applications such as SGLang and vLLM. I find Linux VERY unstable otherwise, though productivity may be higher.

Hardware

RX 9060 XT 16GB. Ryzen 7 5800X with 16GB DDR4 system RAM.

Local models

Qwen3.8 27B fits nicely at around 3โ€“4 BPW on my hardware โ€” a little slow, but it works. Qwen3.6 35B A3B (LuffyTheFox's version) is good for other applications. Currently trialling Qwen3.8 Flash Next.

Do I think AI is good

Very good nowadays, but it needs the right setup and use โ€” otherwise I think a good developer with AI gets work done 2โ€“5ร— faster. You also don't need coding knowledge to use AI to code these days; the latest models are very good, even at the low end.