Design two recommendation systems for a large visual-discovery platform: 1. **Notification recommendation system**: Decide whether to send a notification to a user, which content to include, and when to send it. The goal is to increase long-term engagement without causing notification fatigue, unsubscribes, or spam complaints. 2. **Home-feed ranking system**: Rank candidate visual posts for each user session. The goal is to maximize high-quality engagement while preserving freshness, diversity, safety, and latency requirements. Discuss product goals, success metrics, data and labels, candidate generation, feature engineering, model choices, online serving architecture, experimentation, cold-start handling, exploration, and major failure modes.