Whenever you're ready, walk me through how you'd approach this.
Describe, in detail, the design of a complex system you built at a previous company. Choose a system you personally designed or substantially implemented, preferably one involving machine learning and production-scale data or serving. Explain the requirements, scale, architecture, model choices, data and training pipelines, deployment strategy, evaluation, monitoring, and the tradeoffs behind your decisions. Be prepared to discuss online versus batch components, feature drift, training-serving skew, reliability, and a significant failure or redesign.