Key contributor to DeepSeek's mixture-of-experts work.
Worked on DeepSeek-V2's MoE architecture — part of what made frontier-class capability cheap enough to give away — after research at Alibaba's DAMO Academy.
No books — their work lives in papers, code and talks. The accounts above are the place to follow it.
No freely licensed portrait was available, so a monogram is shown. Roles and dates are drawn from public professional records. Where a current role is listed, it reflects the most recent information available and may have changed.