Image: abhay.fyi
An Oral-B iO5 toothbrush read over Bluetooth and its model distilled
Claude read the Bluetooth stream of an Oral-B iO5, built a live 3D simulator, extracted the app's brushing model and distilled it into a much smaller network.
Credits
Earliest dated source:
The full story
The owner bought an Oral-B iO5 for its 6-axis motion sensor and let Claude loose on it. Within minutes it read the Bluetooth stream and worked out what was what, with the owner pressing buttons while it watched, then built a live three.js simulator.
With a phone in developer mode, Claude pulled the app and extracted Oral-B's model, a 1.2 million parameter bidirectional GRU. It found a public brushing dataset, evaluated the model, found only about 10 seconds of history mattered, and distilled it into tiny models including a CNN 100 times smaller that matched the original. The owner says Claude made only a couple of glaring mistakes.
How it was done
Tools
Themes
Loading themes…
Sources
From a short post by the maker, so the details are as they tell them.
- Reddit: Claude + Oral-B iO5 -- built a live simulator & distilled oral-b's models into a tiny CNNReddit poster: hayAbhayPosted
The maker's post on r/ClaudeAI.
If it disappears: archived copy of Reddit: Claude + Oral-B iO5 -- built a live simulator & distilled oral-b's models into a tiny CNN (opens in a new tab)
- The project: abhay.fyi
Linked in the post.
- Reddit: Streaming Oral-b iO5's motion sensor data + distilling & running its ML model locally + li
The maker's Home Assistant post describing the IMU stream, live 3D twin and model-distillation experiments.