TechFlow reports that on October 4, Vitalik.eth posted on Farcaster revealing that he is conducting a personal experiment combining local models with remote frontier models to generate personalized diet and exercise recommendations using personal health and travel data while preserving privacy.
The system adopts a three-layer privacy protection architecture:
- Identity layer: Local models (Qwen 3.8B) replace users to construct query requests, preventing writing style from exposing identity
- Payment layer: Uses zkAPI to hide payment information
- Network layer: Hides IP addresses via Tor
The system is currently operational, but Vitalik points out three shortcomings: Tor has low efficiency and high latency for per-request disassociation; the local model's processing speed is only 20-30 TPS, while a smooth experience requires 100+ TPS; and the stricter the data protection, the more limited the assistance the remote model can provide. Relevant code has been submitted to the Ethereum zkAPI repository.





