Build the foundational agentic AI layer for a materials-science platform, including multi-model provider abstraction, agent orchestration with stateful checkpoints, retrieval systems, prompt versioning, and comprehensive tracing and evaluation frameworks. You'll design agents that plan and reason over tool calls in production, implement human-in-the-loop safety gates, and ensure all LLM behavior remains auditable and cost-tracked across customers' secure environments. The role demands deep production experience with agentic and LLM systems: async Python, structured outputs, memory and context management, multi-step workflow orchestration, and evaluation harnesses that catch regressions before deployment.