We share some of our work in the open. Each project was built with AI coding tools, and you can see it working: every project has a video or a slide-by-slide journey, and most also publish their source, prompts and full conversation on GitHub. They are meant to help people build applications of their own. Here they are:
A series building tools for narrating and recording on a Mac, starting with your own head on screen.
See how a native macOS app was built from scratch using Claude, in 17 minutes, by voice alone. It shows your webcam as a bubble floating above every app, blurs the room behind you or puts a picture there, and transcribes what you say as you speak.
A series exploring visual modeling, workflow graphs, and discrete-event simulation, using a restaurant as the subject.
See how a working restaurant simulation was built from scratch using Claude. Guests arrive, waiters take orders, cooks work through tickets, and every object on the floor can be inspected.
See how a second view was added to the same simulation using Claude. It shows work flowing through steps and queues, with a measure and a target for everything on screen.
See how the restaurant was rebuilt on discrete-event simulation using Claude. One engine, three views (Floor Plan, Process Map, Scorecard), an objective and metrics for every object, and a full test suite.
A series using AI to plan real occasions, by talking them through.
See how a real tea party was planned by voice, and an app built while doing it, using Claude. A 3D table that turns by itself, finger sandwiches with recipe cards, a drinks table with steam and bubbles, and a plan for the day with a shopping list by store.
A series building a live view of the New York subway from the MTA's public data.
It is an experiment, not an official MTA or NYC Subway product, and should not be relied on for travel.
See how a live subway map was built from scratch using Claude. It covers all 29 lines and 496 stations, shows service alerts as the MTA publishes them, keeps a history of them, and lets you wind the clock back.
An agentic LLM system for practising system design and other technical interviews, by voice.
See a full system design interview with an AI coach. An expert writes the challenge, one agent turns it into a cheat sheet, and another uses it to interview you and grade your answers. Built with Gemini CLI and ChatGPT, and free to try as custom GPTs.
An MCP server and a Lightroom Classic plugin that let an LLM edit RAW photos by voice.
See an AI edit a RAW photo in Lightroom Classic, by voice. An LLM such as Claude, ChatGPT or Kiro moves the sliders itself: tone, curves, colour mixing and grading. Built mostly by voice with Kiro CLI, in Rust with a Lua plugin, and tested to a high standard: more than 300 automated tests and a simulated Lightroom.
Learn tone curves from an AI that edits as it teaches. It explains the master, red, green and blue curves to a photographer and applies each one by voice.
Click the picture for twice the size · move the mouse to look around · Escape to close