NEP (Notification Event Processor) is a real-time notification system developed by Pinterest. It improves user engagement through personalized notifications based on various factors. NEP uses machine learning to determine content, recipients, channels, and optimal timing for notifications. Its key modeling components focus on what to send, to whom, through which channel, at what time, and how to prioritize content. NEP introduces a daily candidate generation approach for notification events, and employs a multi-head model (NEP ranker) to predict user engagement probabilities. The Policy component makes notification send decisions based on utility scores and thresholds, which are automatically tuned using a PID controller to align send volume. NEP's implementation resulted in significant improvements in user email and push engagement metrics, as well as weekly active user (WAU) growth. The system is served in real-time using a GPU cluster and low-precision model serving is planned for future optimization. NEP demonstrates the importance of notifications in driving user retention and highlights the benefits of leveraging machine learning for personalized and timely engagement strategies.