LLM Inference Optimization: How to Make LLM Calls Faster

LLM-powered features are slow by default, and the slowness is user-visible. The teams that ship fast LLM features apply a small set of optimizations -- streaming, prompt caching, smaller models for routing, parallel calls, prefetching. Here's how to actually make LLM calls feel fast.