← All field notes

Field note 03 / 03

2 July 2026
3 min read

Affective computingMaking

The posture model was the easy part

A useful ML prototype depends as much on feedback timing and framing as it does on prediction quality.

Perfect Posture movement-analysis interface

A posture classifier can return an answer in milliseconds. Deciding when that answer helps a person—and when it merely interrupts them—is the more interesting problem.

Perfect Posture treated feedback as part of the model loop. Too frequent and it became noise; too late and the connection between movement and response disappeared.

The strongest prototype was not the one with the most information. It was the one that made the next useful action obvious.

Continue the conversation

A thought worth challenging?

Write to me →