LLMOps

The discipline of operating LLM-based applications after the prototype: versioning prompts and models, observing cost and latency, evaluating quality, testing regressions, and responding to incidents. The operational object is not only the model. It is the combination of data, context, tools, policy, and interface that produces observed behavior.