Minsheng Hao (郝敏升)
I am now an LLM post-training engineer at Tencent, based in Beijing. My work focuses on post-training, agent systems, and evaluation for large-scale foundation models, spanning from the language of life to human natural language.
I received my Ph.D. from Tsinghua University in 2025, advised by Professor Xuegong Zhang. Along the way, I worked in the United States at Genentech’s Aviv Regev Lab, in Beijing at HongShan (Sequoia China), and on foundation-model research with BioMap.
My work has been recognized by AI talent programs including Qingyun, AliStar, TGT, and BeiDou. In parallel, I received the 2025 Ray Wu Prize for Excellence in Life Sciences, and contributed to scFoundation, which was named one of China’s Top 10 Bioinformatics Breakthroughs of 2024.
News
- 2026-04 My Google Scholar citation count crossed 1,000.
- 2026-01 Joined Tencent as an LLM post-training engineer.
- 2025-07 Gave a speech as the graduate representative of the Department of Automation at Tsinghua University.
- 2025-06 Shared PerTurboAgent, an LLM-powered agent for iterative Perturb-seq design developed during my Genentech internship.
- 2025-01 Started at HongShan (Sequoia China), researching agentic AI and contributing to Xbench.
- 2024-07 Worked as an international student intern at Genentech R&D in South San Francisco.
Experience
- Tencent — LLM Post-training Engineer, Beijing, 2026-present.
- HongShan (Sequoia China) — Strategy Analyst Intern, Beijing, 2025. Researched agentic AI for pharmacy and contributed to Xbench.
- Genentech, Aviv Regev Lab — Student Intern, South San Francisco, 2024. Worked on LLM-powered agent systems in a real-world research environment in the United States.
- BioMap — Research on large-scale foundation models for single-cell transcriptomics and AI for biology.
Selected Work
- PerTurboAgent — an LLM-powered agent for iterative Perturb-seq design.
- Large Scale Foundation Model on Single-Cell Transcriptomics — foundation-model research at the intersection of AI and biology.
- scDiffusion — diffusion modeling for conditional generation of single-cell data.
- STEM — transfer learning for integrating single-cell and spatial transcriptomics.
For a full list of publications, please visit my Google Scholar.
Last updated: 2026.04