Laboratory
The lab log
Not a portfolio: a public log of engineering experiments. Each one has a problem, a hypothesis, the technology, what is measured and an honest result
EXP-001 Running
VibeCatter in 3D
- Problem
- The character has to live on the site, not only in videos
- Hypothesis
- A rigged cat model can load after the page is idle and still feel instant
- Technology
- Three.js
- glTF / GLB
- meshopt
- WebP
- GSAP
- Metrics
- 3.9 MBmodel file
- 116 046triangles
- 28facial morphs
- 8ear and tail clips
- Result
- Live on the home page. Without WebGL, with reduced motion or a slow connection the portrait stays instead. Next: the build with 52 ARKit expressions and 12 body clips (talk, wave, think, point)
- Episode
- Not in an episode yet
- Code
- Not published
EXP-002 Graduated
A phone as a gamepad
- Problem
- A live show needs everyone in the room to take part without installing anything
- Hypothesis
- A QR code and a browser are enough to turn any phone into a controller
- Technology
- Socket.IO
- Node.js
- PixiJS
- QR
- Metrics
- 60server updates per second
- 20players in one room
- 0apps to install
- Result
- It worked well enough to become its own product: QRuma. The same pairing will steer the interactive story on the home page
- Episode
- Not in an episode yet
- Code
- Not published
EXP-003 Building
An honest scoreboard
- Problem
- A war against a number needs the number, and the number is rounded
- Hypothesis
- Public channel statistics, labelled by accuracy, make a scoreboard people can trust
- Technology
- YouTube Data API v3
- Next.js
- Metrics
- 3significant figures YouTube keeps
- 60minutes at most another channel’s figures stay in memory
- Result
- The connector is built; it waits for an API key. Until then the War Room shows no figures at all
- Episode
- Not in an episode yet
- Code
- Not published
EXP-004 Planned
The phone as a story remote
- Problem
- A visitor watches the cat; they should be able to answer him
- Hypothesis
- The big screen plays the 3D scene, the phone sends choices, and a deterministic scene graph picks the next scene
- Technology
- Three.js
- GSAP
- Socket.IO
- Scene graph
- Result
- Not started. The metrics will be the share of visitors who connect a phone and the share who finish the main branch
- Episode
- Not in an episode yet
- Code
- Not published
EXP-005 Specification
AI Scene Engine
- Problem
- Every interactive scene is built by hand, and that does not scale to a series
- Hypothesis
- A local browser studio and an animation engine with interfaces for AI agents can assemble scenes from a description
- Technology
- TypeScript
- WebGPU / WebGL
- AI agents
- Result
- A specification exists; nothing is implemented or measured yet
- Episode
- Not in an episode yet
- Code
- Not published
