Build a recommendation engine without tracking users

Collaborative filtering needs user history. Content-based needs user profiles. Apple's on-device ML works but limits cross-device experience. Differential privacy adds noise. Contextual bandits need some session data. The EU's AI Act may classify recommendation systems as "high-risk."

What a strong answer does

• Recommendation quality vs privacy tradeoff
• Technical feasibility
• Novelty of approach
• Regulatory compliance