Flagship project · actively developed

Local AI. No black box.

Hermaeus is a native, local-first AI workspace built to keep models, context, memory, agents, and runtime behaviour visible and under your control.

  • .NET 10
  • Avalonia UI
  • Windows + Linux
  • Local-first
Hermaeus first-run setup screen for choosing local data and AI asset folders
Real UI, local setup, no cloud account bullshit.
LOCAL MODELSREVIEWABLE MEMORYCONTEXT RECEIPTSLOCAL RAGRUNTIME TRACESSUPERVISED AGENTSVOICE WORKFLOWSBENCHMARKINGNATIVE DESKTOPDIAGNOSTICSLOCAL DATA
01Local by defaultLocal models, data, and workflows
02Observable by designContext, citations, traces, and runtime state
03Human in controlSupervised agents and explicit approvals
04Native desktopNo browser shell pretending to be an app

Selected work · Local AI

Start where the systems get interesting.

Hermaeus is the flagship. The work around it explores how local models run, retrieve, observe, and behave when the convenient abstractions are removed.

Engines + simulation

Purrfect Chaos.

A mischievous cat sandbox game built on Chaos Engine: an in-house, MVVM-based game engine in C# and .NET 10. Game code and engine code are growing together.

  • In-house engine
  • Linux and Android targets
  • Keyboard, mouse, and gamepad input
  • Quests, progression, economy, and minigames
Chaos Engine

Current Alpha captures show the game as it exists, rough edges included.

Also: DynoForge simulation

Tools + experiments

Different domains. Same appetite for the hard bit.

Desktop utilities, simulation, audio work, game-engine experiments, and the occasional project that started as a question worth following.

Field notes

The build is the interesting bit.

Technical walk-throughs of the decisions underneath the headline: runtime boundaries, retrieval, engine architecture, and experiments that stay inspectable.

Read the field notes

About

I build the bits worth understanding.

Software engineering · systems · local AI · desktop · experimentation

I build software, break assumptions, and disappear down technical rabbit holes until something useful comes back out.

My work spans local AI, developer tooling, native desktop applications, systems engineering, graphics, and experimental projects. I tend to favour understanding and owning the important parts of a system instead of reaching for another dependency by default.

I care about software that can explain itself. Build it. Use it properly. Measure what happens. Find where the assumptions were wrong. Improve it. Repeat. If something only works in a demo, or cannot survive real use and a second look, it is not done.

I’m comfortable working where the abstractions stop being helpful: performance constraints, native runtimes, file formats, rendering, and model behaviour.

Engineering posture

The site is part of the portfolio.

01

Dependency-light

Semantic HTML, modern CSS, and only the JavaScript that earns its place. No framework and no package-manager supply chain.

02

Built for the browser

Fast, direct, and designed to get out of the way. The work should be doing the talking.

03

Accessible under pressure

Keyboard-first navigation, strong focus states, reduced-motion support, semantic structure, and resilient responsive layouts.