Vitalik: Utilizing local AI models and remote tool calls to generate personalized health recommendations

Vitalik shares his personal experiment, utilizing his own health and travel data, combined with cutting-edge models to generate personalized dietary and exercise recommendations, while avoiding leaking personal information to remote models. The system uses the local model Qwen 3.8 Flash Next for orchestration and calls the remote model as a tool. The system adopts a three-layer privacy protection architecture: the identity layer uses the local model Qwen 3.8B to construct query requests, the payment layer uses zkAPI to hide payment information, and the network layer hides IP addresses through Tor. Currently, the system has successfully run and received suggestions.