Reduce LLM API costs without sacrificing quality. Covers prompt caching, model routing by task complexity, prompt compression, output truncation, and batching. Use when LLM bills are growing or you're scaling to many users.