This talk will provide an overview of best practices in designing and scaling both synchronous and asynchronous machine learning (ML) APIs. This includes optimizing the APIs to deliver the best performance from a latency, accuracy, and cost perspective. The talk will focus on key checklist items before the handoff between from ML teams to engineering, scaling in production including infrastructure management, monitoring/troubleshooting of issues, and ML OPS. Lastly, key learnings from updating ML APIs in production while minimizing user impact will be shared