Vitalik Tests Privacy-Preserving AI Health System: Local Model + zkAPI + Tor Three-Tier Architecture Prevents Identity Leakage

Vitalik.eth disclosed on Farcaster that he is conducting a personal experiment by combining local and remote frontier models to generate personalized dietary and exercise recommendations based on personal health and travel data, all while preserving privacy. The system employs a three-layer privacy protection architecture: • Identity layer: A local model (Qwen 3.8B) constructs query requests on behalf of the user to prevent writing style from exposing identity. • Payment layer: Uses zkAPI to hide payment information. • Network layer: Conceals IP addresses via Tor. The system is currently functional, but Vitalik points out three limitations: Tor is inefficient and incurs high latency for per-request unlinking; the local model operates at only 20-30 TPS, whereas a smooth experience requires 100+ TPS; and the stricter the data protection, the more limited the remote model's assistance. The relevant code has been submitted to the Ethereum zkAPI repository.