PeopleofSuperintelligence

Researcher

Shubham Toshniwal

Also known as शुभम तोषनीवाल

Shubham Toshniwal is an AI researcher whose work spans speech systems, coreference resolution, mathematical reasoning and knowledge agents. In 2018, they co-authored an oral NAACL HLT paper on integrating lexical and acoustic-prosodic information into speech parsing. Their cross-domain coreference work earned the CRAC Best Short Paper Award in 2021; they completed a PhD in entity tracking at TTI Chicago in 2022 and joined FAIR, Meta AI, as a Research Scientist. At NVIDIA, they co-authored OpenMathInstruct-2 and helped build the NemoSkills system that won AIMO-2 in 2025. They are a Senior Research Scientist at Databricks Mosaic, where they co-authored OfficeQA Pro V2 in August 2026 and develop knowledge agents with synthetic data and reinforcement learning.

Works
47
Citations
1,599
h-index
14
Badges

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

  1. 2018

    Speech parsing at NAACL

    In 2018, they co-author “Parsing Speech: A Neural Approach to Integrating Lexical and Acoustic-Prosodic Information,” an oral NAACL HLT paper.

  2. 2021

    A CRAC short-paper award

    They co-author “On Generalization in Coreference Resolution,” which combines several datasets to improve model generalization across domains. In 2021, the paper wins CRAC’s Best Short Paper Award.

  3. 2022

    A PhD in entity tracking

    They complete their PhD at Toyota Technological Institute at Chicago in 2022 with a thesis titled “Efficient and Interpretable Neural Models for Entity Tracking.” Their advisors are Kevin Gimpel and Karen Livescu, and they hold a BTech in computer science from IIT Kanpur.

  4. 2022

    Research Scientist at Meta AI

    After graduating from TTI Chicago, they become a Research Scientist at FAIR, Meta AI, in 2022.

  5. 2024

    OpenMathInstruct-2 dataset

    In 2024, they co-author OpenMathInstruct-2, a math-reasoning dataset with 14 million question-solution pairs and about 600,000 unique questions. This work belongs to their OpenMath series at NVIDIA.

  6. 2025

    AIMO-2 first-place solution

    At NVIDIA, they co-author the NemoSkills solution that wins AIMO-2 in 2025. The system answers 34 of 50 problems and uses the OpenMathReasoning dataset, with 540,000 math problems and 3.2 million long-reasoning solutions.

  7. August 2026

    OfficeQA Pro V2 benchmark

    In August 2026, they co-author OfficeQA Pro V2, a benchmark for evaluating enterprise-style grounded reasoning in AI agents.

  8. As of October 2026

    Senior Research Scientist at Databricks Mosaic

    They are a Senior Research Scientist at Databricks Mosaic in New York City. Their research uses synthetic data and reinforcement learning to train custom knowledge agents, and they contribute to KARL and Instructed-Retriever-1.

Career

  1. NowDatabricksSenior Research Scientist
  2. 2024NVIDIA
  3. From 2022FAIR (Meta)Research Scientist

Education

  1. 2022Toyota Technological Institute at ChicagoPhD
  2. Indian Institute of Technology KanpurBTech in computer science

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

Most-cited works

  1. Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models2022 · 569 citationsWith Abhinav Rastogi, Abubakar Abid, Andrea Madotto, Ashish Sabharwal, Chris Callison-Burch and Christopher D. Manning
  2. Multilingual Speech Recognition with a Single End-to-End Model2018 · IEEE International Conference on Acoustics Speech and Signal Processing · 216 citationsWith Ron J. Weiss and Tara Sainath
  3. Lingvo: a Modular and Scalable Framework for Sequence-to-Sequence Modeling2019 · 169 citationsWith Colin Cherry, Dmitry Lepikhin, Melvin Johnson, Ron J. Weiss, Tara Sainath and Yonghui Wu
  4. A Comparison of Techniques for Language Model Integration in Encoder-Decoder Speech Recognition2018 · IEEE Spoken Language Technology Workshop (SLT) · 149 citationsWith Karen Livescu, Tara Sainath and Yonghui Wu
  5. Multitask Learning with Low-Level Auxiliary Tasks for Encoder-Decoder Based Speech Recognition2017 · Interspeech (USB) · 121 citationsWith Karen Livescu

From OpenAlex, as of 21 October 2026.

  1. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Shubham Toshniwal. 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. 16% of AI career moves in the 2020s went from a university to a companyIncludes Shubham Toshniwal. 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.
  3. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other universityIncludes Shubham Toshniwal. 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.
  4. Among employers with 30+ full-time cases, OpenAI leads primary technical role H-1B pay at $310,000, and Amazon, Google and Microsoft file 57% of all casesIncludes Databricks. Across 440,669 certified US H-1B filings from 44 employers (FY2020–FY2026), OpenAI leads primary technical role base pay at $310,000 and xAI leads overall base pay at $290,000.

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 Shubham Toshniwal yet. Send us a free-licensed photo.

  1. Publications | Shubham Toshniwal shtoshni.github.io. Checked 20 October 2026.
  2. Speech and Language @ TTIC: News slattic.ttic.edu. Checked 20 October 2026.
  3. On Generalization in Coreference Resolution - ACL Anthology aclanthology.org. Checked 20 October 2026.
  4. Efficient and Interpretable Neural Models for Entity Tracking arxiv.org. Checked 20 October 2026.
  5. Shubham Toshniwal shtoshni.github.io. Checked 20 October 2026.
  6. OpenMathInstruct-2: Accelerating AI for Math with Massive Open-Source Instruction Data arxiv.org. Checked 20 October 2026.
  7. AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset arxiv.org. Checked 20 October 2026.
  8. Math Test? No Problems: NVIDIA Team Scores Kaggle Win With Reasoning Model blogs.nvidia.com. Checked 20 October 2026.
  9. Introducing OfficeQA Pro V2: A New Benchmark for Enterprise Grounded-Reasoning databricks.com. Checked 20 October 2026.