Design a personalized **news feed** recommendation system. Requirements: - Low latency serving (real-time feed generation). - Personalization using user behavior and content features. - Freshness and diversity (avoid showing only one topic/source). In your design, explain: 1. High-level architecture (candidate generation, ranking, post-ranking). 2. Data/logging needed for training and evaluation. 3. How you would generate and serve **embeddings** (user/item/content embeddings): - How embeddings are trained. - How they are updated (batch vs near-real-time). - How the online service fetches/uses them (feature store vs embedding store). 4. How you evaluate the system offline and online (A/B tests, metrics, guardrails).