Netflix has centralized its approach to personalization by developing a Foundation Model that consolidates user preferences from extensive interaction histories. This shift aims to streamline the maintenance of various specialized models that previously powered the Netflix homepage, saving time and resources while enhancing overall performance. Instead of relying on multiple models, Netflix now uses this powerful model to distribute learnings across different applications. The integration of the Foundation Model involves three primary approaches: embeddings, subgraph, and fine-tuning. The embeddings approach is the simplest, allowing applications to use embeddings as features for generating recommendations. However, a drawback is the potential staleness of recommendations due to delays in processing. The subgraph method enables a deeper integration, but it also introduces complexities, such as increased application model size and longer inference times. This approach is best reserved for high-impact cases where the benefits outweigh the additional costs. The fine-tuning option allows teams to tailor the Foundation Model to specific needs by adjusting its parameters based on domain-specific data. This is particularly useful for different sections of the Netflix platform, where user interactions vary in importance. Teams can either fine-tune the entire model or freeze certain layers while adding new objectives. A framework has been established to facilitate this fine-tuning process, making it easier for teams to create custom versions of the Foundation Model tailored to their applications.