OpenAI ML System Design interview questions focus on building reliable, scalable systems that run modern machine learning — especially large language models — in production. Expect problems that blend classic system design (APIs, databases, caching, load balancing, observability) with ML-specific concerns such as model training and serving (distributed training, GPU/TPU utilization, batching), token and context management, model routing and versioning, latency vs cost trade-offs, and safety/privacy safeguards. Interviewers evaluate your ability to scope ambiguous problems, state assumptions, justify trade-offs, and communicate a clear, testable architecture rather than a perfect end-to-end spec. For interview preparation, practice designing end-to-end LLM-backed features at multiple scales: from a single-model API to a globally sharded, cost-optimized inference fleet. Emphasize clarity (diagrams and interfaces), metrics and monitoring, failure modes and fallbacks, and data governance. Run timed mock designs that force you to prioritize requirements, call out safety and privacy considerations proactively, and explain why certain ML-specific choices (caching, summarization, routing, batching) matter for both performance and cost.