System design for ML engineers: how different is the interview?

As a data engineer, I'm increasingly seeing system design rounds tailored for ML engineers. How different are these interviews from traditional backend system design? Beyond the usual components like databases, caches, and APIs, what new areas are typically expected? I anticipate topics like model serving infrastructure (online vs. batch inference), feature stores, data pipelines for training (ETL, data versioning), and maybe even MLOps considerations like model monitoring and retraining loops. My concern is balancing the ML-specific components with the foundational distributed systems concepts. How much traditional system design depth (e.g., consistency models, network protocols, load balancing) is still expected versus the ML-specific knowledge? Are there particular resources or frameworks for preparing for these hybrid ML system design interviews? I want to make sure I'm not over-indexing on one area at the expense of another.