evaluated GPT-6: Astra, Sol, Luna in Codex, with a focus on agentic coding, CUA, memory/compaction, connector usage, and reasoning calibration; iterated toward a secure, reliable eval environment; helped incubate an agent-driven evaluation system for R&D.

improved coding agent quality with better context and failure analysis; explored training lower-cost subagents with multi-agent RL.

worked on early-stage LLM development behind initial Apple Intelligence launches, including Writing Tools and Reply Suggestions.

sequential recommendation and retrieval; AutoML