Three constraints that shape every on-prem build
Model selection, infrastructure sizing, and access governance are the three constraints that define an on-prem LLM build. The talk covers how to score open-weights candidates against your own benchmarks, how to size for inference and not training, and how to keep the access controls auditable after the system is live.
Operations is the part that beats the project
A deployment that ships without an operations plan quietly degrades. The recording covers the runbook surface area that keeps a deployment honest in year two: prompt rot detection, evaluation suites that survive model updates, and the governance posture that auditors actually accept.