System Design Article

Design Facebook News Feed

Difficulty: Hard

Design Facebook's News Feed for 2 billion daily active users where every feed open reads from a personalized, ML-ranked timeline assembled from thousands of candidate posts in real time. Unlike Instagram's chronological precomputed feed or TikTok's pure recommendation, Facebook blends a friend graph, group memberships, page follows, and ads into one ranked stream via the legendary EdgeRank-and-successor algorithms. The interview centerpiece is the aggregator pattern: parallel candidate retrieval from many sources, real-time feature lookup, ML scoring, and online filtering, all under a 200 ms p99 budget. We also cover real-time updates (push notifications when a friend posts), edge ranking signals, and how Meta keeps the feed fresh with no precomputed timeline.