Feature stores have enabled several machine learning teams to build more reliable models and iterate faster, including by resolving online-offline inconsistencies and enabling feature reuse. However, most feature stores are designed for tabular data. In this tutorial, Laurel Orr et al. focus instead on managing self-supervised pretrained embeddings as model features. They discuss challenges associated with managing training data, evaluating embedding quality, and monitoring embedding-based models. In concluding, they identify the opportunity to build features stores with native support for embeddings and develop tools to patch errors in embedding-based ML systems.