Local AI. Thoughtful desktop apps. Automation with a purpose.
By day, I work on laboratory software at LabServant.
After hours, I turn curious ideas into things you can actually use.
Four projects, four ways to put more control in your hands.
Speak. Keep working. Local speech recognition, personal corrections, live typing, and optional local rewriting. A desktop voice companion with a living ocean interface.
Electron · React · TypeScript · Python · MLX / CUDA
Explore the project → · Download a release
Open a window into Delulu Talks
Project screenshot from the browser preview with demo text.
Give an agent its own lane. A native KWin plugin and programmable CLI for independent desktop input, reusable actions, and private agent environments on KDE Wayland.
Python · C++ · KWin · Wayland · CLI
Explore the project → · See how it works
A quick peek or a deep investigation. Four research modes—peek, skim, analyze, research—with source-cited answers, evidence graphs, MCP tools, and a terminal interface.
Python · MCP · SearXNG · Local LLMs · TUI
Explore the project → · Read the research contract
Your local models, one workbench. Streaming chat, generation controls, model management, and exportable run history, with a Rust backend talking directly to Ollama.
Tauri · Rust · React · TypeScript · Ollama
Explore the project → · Under the hood
Step inside Model Yard
I work on software that helps laboratories manage their daily operations. My work reaches across:
- Laboratory workflows: equipment, ordering, inventory, reservations, and risk assessment.
- Interface modernization: Laravel and Livewire screens, reusable components, grids, and the details that make a busy interface easier to use.
- The connective tissue: integrations, exports, printing, authorization, and maintaining the bridge between modern and legacy code.
- Developer tooling: repository-aware AI workflows, component guidance, localization, and documentation.
I like working through the whole feature: data and behavior, the interface people touch, and verification in the actual browser.
My recurring rabbit holes: speech models, agent tools, desktop UX, game experiments, and making complicated systems feel approachable.
Side quests & earlier chapters
From PHP CRUD and canvas games to desktop AI and native input systems:
- Garage — early PHP application work.
- bobgame — JavaScript canvas experiments.
- BFGautofarmGT — an early Python automation experiment.
- DeepSeek Harness fork — desktop experimentation with an existing agent harness.
A small signal from my public repositories, refreshed by GitHub Actions.
Have an idea worth a side quest?
Open an issue on the relevant project. I like a good rabbit hole.