At TalentsAI, I design adversarial Python coding-evaluation tasks for frontier LLMs, with specifications, independent reference oracles, and hidden test suites. I validate solutions and grade model-generated code, then use blind multi-model trials to analyze failure modes while keeping each task uniquely solvable.
At Zhejiang University, I trained, tuned, and evaluated a model for an AI application course project, integrating it into the team’s app for a live demo. I also evaluate Chinese AI output and have annotated Chinese-language data for AI projects; I won First Prize in an AI error-correction competition.
China