01 / About

About me

Profile

His work has evolved from search, question answering, and recommendation systems to language intelligence, data intelligence, large models, and agents, with a focus on applying model capabilities to real-world problems. Since 2010, he has worked in R&D and management at Tencent, 360, Beike, and Rushi, spanning algorithm research, systems engineering, technical direction, cross-functional team building, and product delivery.

He is currently a senior algorithm researcher and foundation-model team lead in 360’s Zhibrain Division, working on LLM post-training, safety and trustworthy generation, RAG and document intelligence, and memory and experience mechanisms for agents.

02 / Career

Career

2023.07
present

360 · Zhibrain Division

Senior algorithm researcher

Research and team leadership across foundation models, post-training, safety, trustworthy generation, RAG, memory, and experience systems.

2017.08
2023.07

Beike · Language Intelligence & Search / Rushi Division

Senior manager / Director / Senior director / Center general manager

Work on enterprise search, data intelligence, AI+VR, and intelligent applications, including Xiaobei Assistant and AI House Tour, with cross-functional team building and leadership.

2012.08
2017.08

360 · Search Division

Technical manager

Research and development in question answering, semantic understanding, dialogue, and data intelligence, including building a dialogue system from zero to one and shipping ad CTR prediction.

2010.07
2012.08

Tencent · Online Media Product Technology Department

Automotive product engineering and recommendation systems.

04 / Contributions

Research & industry contributions

Publications & open models

His first-author or corresponding-author publications appear at NeurIPS, ACL, EMNLP, and AAAI. Representative work includes InduOCRBench and BEAR for document RAG, TriPlay-RL and Magic Token for LLM safety alignment, and research on model merging, long-context training, and agent experience transfer. He led the training and release of the open reasoning model TinyR1.

Patents & standards

He holds granted invention patents in LLM safety and model optimization, knowledge bases and corpora, search and question answering, and intelligent interaction. He contributed to three national and industry standards covering large-model evaluation, knowledge-graph frameworks, and competency requirements for VR professionals, as well as the Report on Large Language Model Safety (2024).

Ministry of Industry and Information Technology

Selected nationally funded projects

  1. Industrial public-service platform for large-model engineering technologies and applications
  2. 2023 key technologies for security testing, monitoring, and protection of AI applications for the industrial Internet
  3. 2024 validation system for key security monitoring and detection technologies for next-generation deepfake and intelligent technologies
  4. General AI challenge program: AI risk management and control software