PeopleofSuperintelligence

Researcher

Kai Hu

Also known as 胡凯

Kai Hu is a researcher at Nex-AGI in Shanghai whose work spans document intelligence and language-model agents. They studied physical sciences and computer science at USTC, earned a joint Ph.D. through USTC and Microsoft Research Asia, and coauthored ViBERTgrid, whose paper won the 2021 IAPR/ICDAR Best Paper Award, and KVPFormer. During a DeepSeek internship, they contributed to multimodal models and DeepSeek-R1; at Microsoft CoreAI, they led development of the Azure Layout API. At Nex-AGI, they coauthored Nex-N1 and MM-Doc-R1 as the group released Nex-N2 with coding and agentic capabilities.

Now at
Nex-AGI and Beihang University
Works
359
Badges

Publication numbers from OpenAlex, as of 22 October 2026. Citations and h-index are hidden because this OpenAlex record may be incomplete or mix in other authors.

Timeline

  1. 2014–2019

    Physics and computing at USTC

    They win first prize in Jiangxi's 31st National Physics Contest for High School Students in 2014. From 2015 to 2019, they study physical sciences and computer sciences at the University of Science and Technology of China in Hefei, earning a B.S. and a dual B.Eng. They intern in MSRA's Multimodal Interaction Group from July 2018 to September 2019.

  2. 2019–2024

    A joint USTC–MSRA doctorate

    They begin a Ph.D. in information and communication engineering at USTC through its joint program with Microsoft Research Asia in September 2019 under Qiang Huo and Jun Du, and complete the degree in December 2024. From June 2020 to June 2024, they intern in MSRA's Multimodal Interaction Group, where they work on Microsoft OneOCR and document intelligence. In 2024, they coauthor Detect-Order-Construct, a tree-construction method for analyzing hierarchical document structure.

  3. 2021

    ViBERTgrid wins best paper

    In 2021, they coauthor ViBERTgrid, a multimodal document representation that joins a BERTgrid to a CNN and trains both jointly for key information extraction. The paper wins the IAPR/ICDAR 2021 Best Paper Award.

  4. 2023

    KVPFormer at AAAI 2023

    As first author, they present KVPFormer at AAAI 2023, a question-answering method that treats key entities in form-like document images as questions and predicts their values with a Transformer decoder. It reports state-of-the-art results on FUNSD and XFUND.

  5. 2024–2025

    Research intern at DeepSeek

    From June to November 2024, they are a research intern in DeepSeek's Multimodal LLM Team in Beijing, contributing to DeepSeek-OCR, DeepSeek-VL2, DeepSeek-V3 and DeepSeek-R1. They are among the authors of the December 2024 DeepSeek-VL2 and DeepSeek-V3 papers; the DeepSeek-V3 report describes a model with 671 billion total parameters. Nature publishes a DeepSeek-R1 paper listing them as an author in September 2025; the study shows that pure reinforcement learning can incentivize reasoning in LLMs.

  6. December 2024–May 2025

    Azure Layout API at Microsoft

    In December 2024, they join Microsoft's Multimodal Intelligence Team in CoreAI as a senior research scientist in Beijing. They work on OCR, document intelligence, retrieval-augmented generation and multimodal language models, and lead development of the Azure Layout API through May 2025.

  7. December 2025

    Nex-N1 agentic models

    In December 2025, they coauthor Nex-N1, a Nex-AGI paper on training agentic models through large-scale environment construction.

  8. Since May 2025

    Researcher at Nex-AGI

    Since May 2025, they are a researcher at Nex-AGI in Shanghai, working on LLM agents. In April 2026, they coauthor MM-Doc-R1, an agentic framework for long-document visual question answering through multi-turn reinforcement learning. The group opens Nex-N2, with coding and agentic capabilities, in May 2026.

Career

  1. Since 2025Nex-AGIResearcher
  2. 2024DeepSeekResearch intern
  3. 2024–2025MicrosoftSenior research scientist
  4. 2018–2024Microsoft Research AsiaResearch intern

Education

  1. 2024University of Science and Technology of ChinaPh.D.
  2. 2024Microsoft Research AsiaPhD
  3. 2019University of Science and Technology of ChinaB.S. and B.Eng.

From the biography.

  1. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Kai Hu. The median h-index is 48 at universities and 14 at frontier AI labs. We compare publishing careers, recent citations, dual appointments and job moves across 2,420 university researchers and 308 frontier-lab researchers.
  2. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other universityIncludes Kai Hu. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other university. We compare doctorate and 5+ year employment links, career timing and current workplaces for 187 major-prize and 1,111 best-paper winners.
  3. 23% of people shaping AI who left a university for a company later returned to a universityIncludes Kai Hu. We estimate that 468 of the 2,011 people shaping AI who left a university for a company (23%) later returned to a university after a median of 5 years, and 163 of 829 (20%) made the reverse round trip.
  4. 16% of AI career moves in the 2020s went from a university to a companyIncludes Kai Hu. We estimate that 16% of AI career moves in the 2020s went from a university to a company; 29% of moves starting at a university went to a company, and 76% ended the university role.

Colleagues

People with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Kai Hu yet. Send us a free-licensed photo.

  1. Kai Hu - Homepage kaihu-kh.github.io. Checked 21 October 2026.
  2. Detect-Order-Construct: A Tree Construction based Approach for Hierarchical Document Structure Analysis arxiv.org. Checked 21 October 2026.
  3. ViBERTgrid: A Jointly Trained Multi-Modal 2D Document Representation for Key Information Extraction from Documents arxiv.org. Checked 21 October 2026.
  4. ICDAR 2021 Awards | ICDAR 2021 iapr.org. Checked 21 October 2026.
  5. A Question-Answering Approach to Key Value Pair Extraction from Form-like Document Images arxiv.org. Checked 21 October 2026.
  6. DeepSeek-VL2: Mixture-of-Experts Vision-Language Models for Advanced Multimodal Understanding arxiv.org. Checked 21 October 2026.
  7. DeepSeek-V3 Technical Report arxiv.org. Checked 21 October 2026.
  8. DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning | Nature nature.com. Checked 21 October 2026.
  9. Nex-N1: Agentic Models Trained via a Unified Ecosystem for Large-Scale Environment Construction arxiv.org. Checked 21 October 2026.
  10. MM-Doc-R1: Training Agents for Long Document Visual Question Answering through Multi-turn Reinforcement Learning arxiv.org. Checked 21 October 2026.