How do you design multi-objective optimization in LLM systems?

Multi-objective optimization balances accuracy, cost, latency, safety, and user satisfaction in LLM systems.

LLM systems are never optimized for a single metric. Instead, they operate under trade-offs between cost, latency, accuracy, and safety. Multi-objective optimization uses weighted scoring systems or dynamic policies to choose models, prompts, or retrieval strategies depending on context and user importance tier.