Shipped applications on large language model APIs, including tool use, structured outputs, retrieval, and multi-step agents, with working knowledge of how these systems fail (hallucination, prompt injection, tool misuse, drift) and how to measure them. The work today centers on AI systems in regulated environments: agentic workflows and automations; the API, authorization, and fraud controls that make them safe; evaluation of systems whose outputs are non-deterministic; and the governance that lets them run in production.