PeopleofSuperintelligence

Researcher

Ivan Sorokin

Ivan Sorokin is a Senior LLM Technologist at NVIDIA and a Kaggle Grandmaster who builds AI systems for math and abstract reasoning. His early research included DARQN, an attention-based extension of DeepMind's Deep Q-Network presented at a NIPS 2015 workshop, and speech recognition and diarization systems built with the Speech Technology Center team for the CHiME challenges in 2018 and 2020. At NVIDIA, he helped win the Amazon KDD Cup 2024 and the AIMO-2 math olympiad competition in April 2025, earned the Kaggle Grandmaster title, and co-authored the papers behind the OpenMathReasoning dataset and the GenSelect method. In December 2025, he and Jean-Francois Puget took first place in the Kaggle ARC Prize 2025 as team NVARC.

Timeline

  1. 2015

    The DARQN attention agent

    He is the first author of Deep Attention Recurrent Q-Network (DARQN), presented at the Deep Reinforcement Learning Workshop at NIPS 2015. DARQN extends DeepMind's Deep Q-Network with soft and hard attention, and in tests on several Atari 2600 games it scores above DQN.

  2. 2018

    The STC system for CHiME 2018

    In 2018, he co-authors the Speech Technology Center (STC) system for the fifth CHiME challenge, on distant multi-microphone speech recognition in homes. The system places 3rd in the single-array track and 4th in the multiple-array track.

  3. 2020

    Target-speaker voice activity detection

    In 2020, he co-authors the STC systems for the CHiME-6 challenge, which covers multi-speaker speech recognition and diarization at dinner parties. For diarization, the team proposes Target-Speaker Voice Activity Detection (TS-VAD), which predicts each speaker's activity in every time frame; the paper is accepted to Interspeech 2020.

  4. July 2024

    Sweeping the KDD Cup 2024

    In July 2024, he and five NVIDIA colleagues, including Chris Deotte and Gilberto Titericz, take first place overall and in every track of the Amazon KDD Cup 2024, a multi-task online shopping challenge for LLMs. With only 96 example questions released, the team builds its own training set and fine-tunes Qwen2-72B-Instruct.

  5. April 2025

    Winning AIMO-2 with NemoSkills

    In April 2025, he is one of seven members of NemoSkills, the NVIDIA team that wins the AI Mathematical Olympiad Progress Prize 2 (AIMO-2) on Kaggle against over 2,200 teams, solving 34 of 50 problems in five hours on four NVIDIA L4 GPUs. Based in Finland, he earns the Kaggle Grandmaster title after the win, and the team directs its $262,144 prize to the NVIDIA Foundation.

  6. April–July 2025

    OpenMathReasoning and GenSelect

    In April 2025, he co-authors the AIMO-2 winning-solution paper, which releases the OpenMathReasoning dataset of 540,000 math problems with 3.2 million long-reasoning solutions, along with code and models. In July 2025, he co-authors GenSelect, a method in which a reasoning model compares N candidate solutions and selects the best one.

  7. December 2025

    First place in ARC Prize 2025

    In December 2025, he and Jean-Francois Puget, competing as NVARC, take first place on the Kaggle ARC Prize 2025 public leaderboard with a 27.64% score on the ARC-AGI-2 benchmark. Their fine-tuned 4B model outperforms far larger models on that benchmark at about 20 cents per task.

  8. As of October 2026

    Senior LLM technologist at NVIDIA

    He is a Senior LLM Technologist at NVIDIA and a member of the Kaggle Grandmasters of NVIDIA (KGMoN), a group that also includes Christof Henkel, Darragh Hanley and Jean-Francois Puget.

  1. Google had the most people shaping AI every year from 2014 to 2025Our estimates of which organizations employed the most people shaping AI from 1961 to 2025, with a year-by-year chart.
  2. The United States employs an estimated 53% of people in our data with a known employment countryOur estimates of where AI careers are concentrated now and how countries’ shares of dated positions changed over time.
  3. 46% of living people in our data with known birth and employment countries work outside their birth countryOur estimate of cross-border careers in AI, from country of birth and the country of each current organization.
  4. More people shaping AI studied at Massachusetts Institute of Technology than at any other institutionOur estimates of where the people shaping AI studied and where their careers led, from education records in cited biographies.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

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

  1. [1512.01693] Deep Attention Recurrent Q-Network arxiv.org. Checked 19 October 2026.
  2. ISCA Archive - The STC System for the CHiME 2018 Challenge isca-archive.org. Checked 19 October 2026.
  3. ISCA Archive - The STC System for the CHiME-6 Challenge isca-archive.org. Checked 19 October 2026.
  4. [2005.07272] Target-Speaker Voice Activity Detection: a Novel Approach for Multi-Speaker Diarization in a Dinner Party Scenario arxiv.org. Checked 19 October 2026.
  5. NVIDIA’s AI Masters Sweep KDD Cup 2024 Data Science Competition blogs.nvidia.com. Checked 19 October 2026.
  6. [2408.04658] Winning Amazon KDD Cup'24 arxiv.org. Checked 19 October 2026.
  7. Math Test? No Problems: NVIDIA Team Scores Kaggle Win With Reasoning Model blogs.nvidia.com. Checked 19 October 2026.
  8. [2504.16891] AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset arxiv.org. Checked 19 October 2026.
  9. [2507.17797] GenSelect: A Generative Approach to Best-of-N arxiv.org. Checked 19 October 2026.
  10. NVIDIA Kaggle Grandmasters Win Artificial General Intelligence Competition developer.nvidia.com. Checked 19 October 2026.
  11. Meet The Kaggle Grandmasters (KGMoN) Team nvidia.com. Checked 19 October 2026.