Whenever you're ready, walk me through how you'd approach this.
Walk me through how you would scale a database horizontally.
Explain how you would evolve from a single database to a distributed system for machine learning workloads. Cover workload characterization, partitioning and replication, routing, consistency, rebalancing, operational safety, and observability. Discuss how your design changes for feature retrieval, training-data ingestion, and low-latency model serving. State assumptions explicitly and call out tradeoffs.