Whenever you're ready, walk me through how you'd approach this.
What is your experience with high-availability systems and how do you ensure their reliability?
Discuss a concrete system you have worked on, including its availability objectives, failure domains, redundancy model, deployment process, observability, and disaster-recovery approach. Explain how you would apply those principles to an ML-serving system, including model, feature, and dependency failures.