Schneider Electric built its LLMOps infrastructure on a self-hosted LangSmith deployment to manage more than 60 AI agents across the organization. The architecture rests on three pillars: observability via workspace-per-product organization, evaluation through offline testing and SME annotation queues, and deployment via dedicated per-product LangSmith Agent Server runtimes on AWS and Azure. The company also built an LLMOps maturity framework tracking instrumentation and evaluation coverage across products, supporting an AI Hub used by 160,000 employees under strict data-residency and security requirements.
