Proven track record building and evaluating production agentic systems on LLMs - multi-step tool use, structured output, multi-agent orchestration, and designing around non-deterministic behavior; ability to build eval datasets from real business data, separate training and blind holdout sets, and catch regressions when models, prompts, or data change. Experience with Recursive Self-Improvement (RSI) for AI agents - building closed loops in which an agent''s own runs, failures, and eval results are fed back (often through another model) to propose, test, and adopt improvements to its prompts, plans, tools, or skills, with rigor to keep them honest: blind evaluation, overfitting audits, regression gates, and human review before production.