Technical
The project is in development. Measurements and quantities on this page are indicative and will be fixed on the prototype.
Hardware
- Tape
- Two-colour flip-disc style cells: a small electromagnet turns a disc that is white on one side and black on the other. Total length about 3 to 4 m. The number of cells depends on the cell size chosen.
- Head
- A carriage on a linear rail, moved by a stepper motor through a belt or lead screw. A limit switch lets the head return to its starting position.
- Control
- An ESP32 microcontroller drives the motor and the cells and receives commands from the computer.
- Camera
- A fixed camera framing the whole tape, or the area around the head.
- Computer
- Runs the controller and handles the language model, the speech synthesis and the sonification.
- Audio
- Speakers for the AI's voice and the sonification. The sound of the cells is acoustic and does not go through the speakers.
Software
- Controller
- Holds the transition table, executes the correct move, compares it with the move the AI announced and writes the step log.
- LLM
- Receives the image of the tape and announces the move. When the controller flags an error, it writes the justification. At the end of the session it writes the story. It can be a cloud model or a local one.
- Max/MSP
- Reads the log and generates the sound using the mapping described in Story and sound.
- Archive
- Stores the story and the log of every session.
Two versions
Minimal version
A shorter stretch of tape that fits on a table. One camera, a laptop, a pair of speakers. It is meant for testing and for small rooms. The logic and the archive are the same as in the full version.
Exhibition version
The full tape of about 3 to 4 m on its own structure, with the rail visible. A sound system chosen for the room, and a local model ready as a fallback if there is no network. A screen or reading station for the archived stories can be added.