Event Discovery and Recommendation System
Design a high-scale event recommendation system for a platform similar to Eventbrite. The system must handle a corpus of millions of ephemeral, location-dependent events and 50M+ users. Detail the end-to-end ML lifecycle, focusing on how you would address the cold-start problem for new events, ensure low-latency geospatial retrieval, and rank candidates based on booking probability. Discuss the trade-offs between model complexity and inference speed, and explain your strategy for handling delayed feedback and temporal data leakage in your training pipeline.
Two-TowerLightGBMFAISSKafkaFlinkSparkSBERTH3RedisMMoE
00