About

A one-person workshop. The work spans full-stack products, research tools, hardware, and design systems — usually in that order, sometimes all at once.

What follows is the trail: shipped projects, archived ones, and the things still being figured out.


Selected Builds

Tools, toys, and explorations.

2025

Handwrite

Full-Stack Product
View Project
AI Ops
Google Gemini 3 ProClaude Sonnet 4.5Claude Opus 4.5
Stack
Next.js 15TypeScriptTailwind CSSPrisma / PostgreSQLStripePythonG-code / Pen Plotter
2025

VibeEcho

iOS App
View Project
AI Ops
GPT-4oGemini 1.5 Flash (multimodal video)
Stack
SwiftSwiftUIFirebaseApifyOpenAI APIGemini APIStoreKit 2
2025

AnimalTaskSim

Research / Open Source
View Project
AI Ops
Claude Sonnet 4.5Gemini 2.5 Pro
Stack
PythonGymnasiumPyTorchStable-Baselines3NumPySciPyPandasMatplotlibPydantic
2026

LubaUI

Open Source / Design System
View Project
AI Ops
Claude Sonnet 4.5Claude Opus 4.5
Stack
SwiftSwiftUISwift Package ManagerMCP
2025 · Archived

Automation & Agents

Workflows
View Project
AI Ops
GPT-4oClaude 3.5 Sonnet
Stack
n8nPythonWordPress API
2025 · Archived

PMX CLI

Developer Tools
View Project
AI Ops
GPT-5 MiniGemini 3 Pro
Stack
TypeScriptNode.jsReact (Ink)OpenAI API

Log

A running record of things built, tested, and changed.

  1. Rewrote the LubaUI page around the 0.2.0 release, and refreshed the agent reference it serves at /lubaui/llms.txt.
  2. Shipped LubaUI 0.2.0 — themes that repaint their components, Reduce Motion honored system-wide, a type scale that grows with Dynamic Type, and built-in strings in English and German. Two of the 104 tests exist only to fail if that wiring ever comes loose.
  3. Started building a real iOS app on top of LubaUI — the first proper test of the system outside its own showcase.
  4. Refreshed AnimalTaskSim around its clearest idea: a virtual lab for testing why animals make the choices they do.
  5. Added a pen-evaluation card to Handwrite: a small tool for comparing how recorded fonts behave on real plotter hardware.
  6. Prepared a VibeEcho release with a corrected in-app credit flow, tested end to end on real iPhone and iPad devices.
  7. Added a hybrid noise-floor check to AnimalTaskSim, making it easier to tell a real behavioral signal from expected model variation.
  8. Expanded AnimalTaskSim’s mouse reference from a small sample to a larger public IBL dataset, while preserving the original comparison set.
  9. Merged the hero with the intro paragraph and bumped the headline. Added a NOW box and this Log.
  10. Rewrote AnimalTaskSim around an adaptive-control agent. Three-circuit architecture, IBL mouse benchmarks.
  11. Shipped the Handwrite project page. Stripe payments, pen plotter pipeline, automated fulfillment.
  12. Site audit pass — dark mode parity, accessibility, performance cleanup.
  13. Rebuilt the Projects page with a terminal-style layout.