Design that gets
built.

WellsDone

A private family organizer for the household — to-dos and boards, a shared calendar, repeating chores, a meal plan, a grocery list that sorts itself by aisle, and house notes — synced live across everyone's phones and computers. It began as a fork of the JourneyLive demo (below), so it keeps that app's layout, drag-and-drop, right-click menus, and quick-add shorthand.

It's designed to stay out of the way. There are no points, scores, or leaderboards — a chore is just done or not done. Parents can change anything; kids check off their own chores and add to-dos and groceries. Each device remembers who's using it, and switching to a parent profile asks for a PIN. A saved meal can push its ingredients straight onto the grocery list, and house notes can be marked parents-only, down to individual lines.

The whole front end is one HTML file, served by a small dependency-free Node server that saves state to disk with daily backups. It runs on a home PC reachable only over a private Tailscale network, and the same API can also run as a Cloudflare Worker with a D1 database behind Cloudflare Access, so nothing has to stay powered on at home.

JourneyLive

A staff-operations app pitched to a church: tasks, a calendar with room booking, a time clock, and request forms that turn into tasks — designed to connect to the organization's Google Workspace. The repository holds a clickable demo, page mockups, and a full build spec for the production version; the production app itself hasn't been built yet.

The demo is a single self-contained HTML file running on sample data. Tasks work like Todoist-style projects with sections, shown as a board or a list, with priority as a color, an “On fire” status that pins emergencies, smart views (Today, Upcoming, My tasks), and quick-add shorthand like Order chairs !1 @Mark #Facilities fri. The calendar blocks double-bookings, and approving an event request builds a whole project from a template. Everything supports drag-and-drop and a site-wide right-click menu, with role-based permissions across eight roles and a dark theme checked against WCAG contrast.

The planned production stack is a TypeScript monorepo — Fastify and Zod on the backend, Prisma on PostgreSQL, and a React + Vite + Tailwind PWA — with Google sign-in limited to the organization's domain. For the pitch, the demo runs on a Cloudflare Worker with a D1 database so every viewer shares one live copy of the sample data.

Plantain

A lightweight, resizable companion mixer for Voicemeeter (Banana and Potato) on Windows — small enough to sit beside a music player, capable enough to replace the full Voicemeeter window for everyday mixing. It's a fork of shuperj's open-source MiniMeeter, redesigned and extended here; all credit for the original concept and groundwork belongs to the original author.

The changes in this fork include a Windows acrylic-glass look that follows the system accent color, a switch between vertical faders and compact knobs with a proportional sizing system so controls never clip at any window size, and edition detection that adapts the strip and bus layout to Banana or Potato. It also adds one-click reverb and delay macros, a karaoke mode cycle, animated audio-reactive backgrounds, A1 output switching from the titlebar, global mute hotkeys, window-size presets, and autostart.

Under the hood it's a Tauri 2 app: a Rust backend loads VB-Audio's Voicemeeter Remote API at runtime and syncs about thirty times a second, so changes made on hardware show up instantly, while the interface is React 19 and TypeScript. Licensed GPL-3.0, like the upstream project.

Overlay Control Hub

The backend for the beanjahmean Twitch stream overlay kit — a local Node.js server (built-in http + ws, no framework) that runs on the streamer's own PC, serves a Windows-95-styled control panel, and pushes live state to every connected overlay over WebSocket.

It does two jobs. First, a Twitch watcher: an anonymous IRC connection flags first-time chatters with no auth needed, while Twitch EventSub (also over WebSocket) catches subs, resubs, and gift subs, each broadcast as an alert with tiered visual/sound effects — Blue/Silver/Gold by sub tier, full RainbowFX for resubs and gifts. Second, a control surface: a plugin-style module system (each module in its own file under modules/, auto-loaded at startup) owning one slice of state — a pushup counter, per-overlay visibility toggles, shared theme colors, four saveable presets, a sticky-note reminder field, and a live design switcher — all persisted to state.json so a restart never loses the pushup count.

The newest piece is that design switcher: the overlay kit ships three full alternate visual themes (Windows 95, Windows XP Luna, and a modern glassmorphism "Minimal" style) sharing the same CSS variable names, served at canonical /overlay/<name>.html URLs. Flipping themes in design.exe broadcasts a state change and every already-open OBS Browser Source reloads into the new look instantly — no re-editing OBS sources, and a fourth theme is just a new folder under themes/ with matching filenames.

Caption Speaker Editor

A standalone local browser app for correcting word-level captions on a short-form video while watching it play back — instead of guessing speaker attribution and timing from a flat transcript file.

Point it at a video file and its caption cards (a cards.json export), and optionally an existing speakers config to reuse the same names, keys, and colors across jobs. It runs a tiny local server (server.py) and opens a browser UI where each caption card can be re-timed, reassigned to the correct speaker, or corrected in place while scrubbing the actual footage — the fix that's tedious to do blind in a text editor.

Video Editor Pipeline

An end-to-end pipeline that takes raw talking-head/stream footage to a captioned, edited DaVinci Resolve deliverable, built around a set of Claude Code skills instead of one monolithic script.

Six stages: Intake organizes raw footage into a per-job folder structure, Transcribe runs WhisperX for word-level timing, Cut identifies the usable ranges, Resolve replays the edit into DaVinci Resolve over its MCP integration, Graphics lays in ProRes overlays via HyperFrames, and Finish renders the deliverable. Each stage is its own skill under .claude/skills/, so a job can be picked up or re-run at any stage rather than starting over.