Experience
- Google DeepMindResearch Scientist — audio LLMs, post-training, controllability
- Meta GenAIEnhanced reasoning with stepwise feedback and long-form reasoning
- NVIDIA ResearchError correction methods for multimodal language models
- Amazon Alexa AIFactuality evaluation agent and synthetic data techniques
Selected Publications
Open Source
Traditional Chinese LLMs
A previous project — open-weight large language models for Traditional Chinese, covering pretraining data, instruction tuning, and evaluation. Saw wide adoption across the research community and industry.
Mistral-Small-Reasoning
Open-weight reasoning model distilled for efficient step-by-step problem solving.