Understanding of LLM internals, including transformer architectures, tokenization, context window management, RAG patterns, prompt engineering techniques, and the security risks inherent to each (e.g., prompt injection, data leakage through context, output manipulation). Engineer observability and visibility into our AI usage and spend, capturing platform logs, building usage pipelines and spend dashboards, and alerting that land in the tools teams already use (e.g., Datadog, Snowflake, Mode).