Building practical AI tools — pointing AI at real problems.

The Living Desktop is a custom interactive desktop environment that replaces a static wallpaper with a live, clickable scene — a row of retro arcade cabinets running real gameplay. It started as a Matrix-style wallpaper with a Spotify lyrics overlay and evolved, version by version, into a working arcade: click a cabinet's coin slot and a Sega Genesis emulator boots a real, fully playable game right on the desktop. ChatGPT generated the cabinet artwork, Claude Code wrote the Python that embeds the video players, wires the audio, and integrates the emulator, and Python holds the whole scene together as one live desktop layer.

TLDR turns a wall of text into the gist, at a length you control. Paste raw text or a web page URL, drag a slider from 1 to 10 sentences, and get a clean summary you can copy or download — no account needed. It's built on Next.js with all AI calls running server-side through Claude Haiku, and pages are fetched and stripped to readable text with Mozilla Readability before summarizing. Because it's a public tool hitting a paid API, it ships with real guardrails: a hard input cap, per-IP rate limiting, and a conservative token budget so it can run unattended without a runaway bill.

AI made words-into-code effortless, but left no good reverse path — people now ship code they can't read. The Code Explainer fixes that by putting the explanation next to the line it describes: your code stays untouched on the left, plain-language margin notes sit beside the relevant lines on the right. A depth toggle switches between Explain (what the code does and why) and Review (risks, smells, edge cases), and it annotates only the load-bearing logic — three sharp notes, not twelve noisy ones. You can export the snippet with the notes injected as inline comments in the right syntax. Built on Next.js with Claude Haiku server-side, language auto-detection, and the same cost/abuse guardrails as the Summarizer.

EDITH is a voice-first AI assistant built to be a genuine working partner, not another chat window — the JARVIS-to-Tony-Stark dynamic. She runs on the desktop with real system agency: shell, screenshots, file access, and browser control by voice, fronted by a custom oxblood-and-bone “aperture” HUD that animates through listening, working, and insight states. The build is phased — from rebranding an existing working program, to a cloud brain on Fly.io, an iPhone client, and long-term memory — targeting sub-2-second voice latency for under $60/month. Deliberately a capable tool, not a companion.

PyQuest is a Mimo-style app for learning Python through short, gamified lessons, with real Python executing in the browser via Pyodide. Ten units and ~28 lessons cover the beginner foundations across five exercise types, including write-and-run code challenges that auto-check your work. XP, levels, daily streaks, and achievement badges keep it sticky; the lessons live in a plain text file you own and can edit.

Tideglow is a cozy underwater life-sim built in GameMaker under the Yardyprince Gaming Co. banner — roughly 90k lines of code and hundreds of hours of design. It blends relaxed fishing-and-farming loops with Zelda-style progression and a darker mystery beneath the surface. Currently in vertical-slice polish: most foundational systems are stable, with content scale-up and late-game beats remaining.

A Godot 4 remake and remix of Shining Force 2, the Sega Genesis SRPG: grid-based tactics battles with full-screen “cut-in” attack and spell cutscenes, overworld town and field exploration, and an SF2-style promotion system where each character promotes once along one of two branching paths. Sprites and animations come through a custom AI-assisted art pipeline.

Agentic OS is the design for a thin integration layer tying a voice assistant (Edith), a personal AI OS (Praetor), Claude Code, and an intel agent into one operating system — speak a goal, it gets classified, routed to the right specialist, tracked, and reported back. The connective substrate is a shared vault every agent already sees.

Castle is the canonical reference for a multi-agent ecosystem: a set of specialized agents, each with its own PRD, organized as a coherent “realm” rather than one monolith. It defines the boundaries, naming, and responsibilities for how the agents cooperate.