PeopleofSuperintelligence

Researcher

Sanjeev Satheesh

Sanjeev Satheesh is an AI researcher whose work spans computer vision, speech recognition and language-model training. He coauthored work on label trees, the ImageNet challenge and end-to-end speech systems, then studied bias in production speech models. At NVIDIA, he contributed to Nemotron-4, Mistral-NeMo-Minitron 8B, Nemotron-H and Nemotron-MIND. As of October 2026, he coauthors an open midtraining recipe for agentic post-training.

Works
38
Citations
44,466
h-index
16
Badges

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

  1. 2011–2014

    Label trees and Deep Speech

    In 2011, he coauthors Fast and Balanced, which learns a label tree for large-scale image classification. In 2014, he coauthors Deep Speech, an end-to-end speech-recognition system.

  2. 2015

    The ImageNet challenge paper

    In 2015, he coauthors the ImageNet Large Scale Visual Recognition Challenge paper.

  3. 2016

    Deep Speech 2 in two languages

    In 2016, he coauthors Deep Speech 2, an end-to-end system for English and Mandarin speech recognition.

  4. 2017

    Reducing production speech bias

    In 2017, he coauthors Reducing Bias in Production Speech Models, which finds that production latency constraints expose bias in end-to-end speech models. Its authors use improved streaming-inference architectures and report error-rate reductions of up to 20% with methods practical for deployment.

  5. 2024

    Nemotron-4 340B models

    In 2024, he coauthors NVIDIA's Nemotron-4 340B Technical Report, which releases base, instruct and reward model versions. The authors also open-source the synthetic-data generation pipeline used for model alignment.

  6. 2024–2025

    Minitron and Nemotron-H

    In 2024, he contributes to the Mistral-NeMo-Minitron 8B foundation-model work at NVIDIA. In March 2025, NVIDIA's Nemotron-H project lists him as a core contributor to evaluations and safety; its hybrid Mamba-Transformer models aim to improve inference speed.

  7. April 2025

    Nemotron-MIND mathematical dialogues

    In April 2025, he coauthors NVIDIA's Nemotron-MIND, a method that turns mathematical content into multi-turn dialogue for pretraining. Its dataset contains more than 138 billion tokens of structured mathematical dialogue, and the project reports gains in math and general reasoning.

  8. As of October 2026

    Midtraining for agentic models

    As of October 2026, he coauthors an open midtraining recipe that prepares base models for agentic post-training. The recipe trains at 256K context on software-engineering data and reports gains on SWE-bench Verified and DeepSWE.

Career

  1. Until 2025NVIDIA
  2. 2016–2018Baidu
  3. 2011–2015Stanford University

From the biography and OpenAlex affiliations (the years of papers that list the organization), so some years are approximate.

Most-cited works

  1. ImageNet Large Scale Visual Recognition Challenge2015 · International Journal of Computer Vision · 38,923 citationsWith Aditya Khosla, Alexander Berg, Andrej Karpathy, Fei-Fei Li, Jia Deng and Jonathan Krause
  2. Deep Speech 2: End-to-End Speech Recognition in English and Mandarin2015 · 2,110 citationsWith Bryan Catanzaro, Dani Yogatama, Sharan Narang and Sherjil Ozair
  3. Deep Speech: Scaling up end-to-end speech recognition2014 · 1,490 citationsWith Bryan Catanzaro
  4. Text Detection and Character Recognition in Scene Images with Unsupervised Feature Learning2011 · Proceedings of the International Conference on Document Analysis and Recognition · 382 citationsWith Andrew Ng
  5. Cold Fusion: Training Seq2Seq Models Together with Language Models2018 · Interspeech (USB) · 279 citations

From OpenAlex, as of 21 October 2026.

  1. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Sanjeev Satheesh. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.
  2. 20% of people at companies who shape AI also hold a university role, led by Google and Stanford UniversityIncludes Stanford University. We estimate that 539 of the 6,766 people with a known current organization (8%) hold both a university role and a company role, including 20% of people at companies and 13% of people at universities.
  3. Google had the most people shaping AI every year from 2014 to 2025Includes Stanford University. Our estimates of which organizations employed the most people shaping AI each year from 1955 to 2025, tracing the shift from university labs to companies.
  4. More people shaping AI studied at Massachusetts Institute of Technology than at any other institutionIncludes Stanford University. We estimate where the people shaping AI studied and where their careers led. Massachusetts Institute of Technology leads with 467 former students, including 291 recorded doctorates.

Colleagues

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

Sources and credits

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

  1. Fast and Balanced: Efficient Label Tree Learning for Large Scale Object Recognition papers.neurips.cc. Checked 21 October 2026.
  2. Deep Speech: Scaling up end-to-end speech recognition arxiv.org. Checked 21 October 2026.
  3. ImageNet Large Scale Visual Recognition Challenge 2015 (ILSVRC2015) image-net.org. Checked 21 October 2026.
  4. Deep Speech 2 : End-to-End Speech Recognition in English and Mandarin proceedings.mlr.press. Checked 21 October 2026.
  5. Reducing Bias in Production Speech Models arxiv.org. Checked 21 October 2026.
  6. Nemotron-4 340B Technical Report arxiv.org. Checked 21 October 2026.
  7. Mistral-NeMo-Minitron 8B Model Delivers Unparalleled Accuracy developer.nvidia.com. Checked 21 October 2026.
  8. Nemotron-H: A Family of Accurate, Efficient Hybrid Mamba-Transformer Models research.nvidia.com. Checked 21 October 2026.
  9. Nemotron-MIND: Math Informed syNthetic Dialogues for Pretraining LLMs research.nvidia.com. Checked 21 October 2026.
  10. Priming Base Models for Agentic Post-Training: An Open Midtraining Recipe arxiv.org. Checked 21 October 2026.