MLOps

The operational discipline of running LLM-based applications in production — prompt versioning, model routing, evaluation pipelines, cost governance, observability, incident response. Distinct from traditional MLOps because model weights are largely opaque and prompts are the primary "code".