Due to competing priorities, getting the first "baby" model out the door on ExecuTorch took far longer than anticipated.
The model monitors a webcam powered by Thingino and served to HomeKit via Scrypted. It produces two primary predictions published to MQTT, which are then exported to HomeKit via Homebridge using mqtt-things:
I have probably bit off more than I can chew with this project. However, the progress made on automated labeling and model conversion has been a game-changer for my local-first AI experiments.
The monetization strategy for private AI in the home is still in its infancy. Do we monetize the orchestration server for private re-training? Or is the value in the "Private Models" themselves? Alternatively, perhaps a data-privacy-first exchange allows for anonymized datasets to be contributed back to the collective in exchange for lower hardware overhead.
A continuation of the previous blog post.
This setup is for educational purposes and reflects a specific point-in-time configuration. Always verify current security practices before deploying to a production-like environment.