Optimise model quality, latency, throughput, tokens, context, retrieval, batching, caching, hardware, routing, and spend.
