PeopleofSuperintelligence

Researcher

Shweti Mahajan

Shweti Mahajan is a Senior Research Software Engineer in Microsoft Research's Redmond Data Science group whose projects span neural networks, language models and generative AI. In 2021, she coauthored studies on training-free geometric neural networks and ranking with semantically similar queries. She coauthored Lexi for user-interface screens in 2022, Orca 2 for small-model reasoning in 2023, and work on contrastive post-training and synthetic data in 2024. In September 2026, she coauthored AI Night-Scientist, an agentic reinforcement-learning framework for scientific ideation. Before joining Microsoft Research, she earned a master's in computer science from the University of California, Riverside.

Works
8
Citations
39
h-index
3

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

  1. 2021

    Learning without gradient descent

    In 2021, she coauthors a study that uses geometrically constrained neural networks to encode and classify MNIST images without any training.

  2. 2021

    Ranking with similar queries

    In 2021, she coauthors a study of ranking with semantically similar queries. The method combines item representations and uses information from similar queries to improve rankings when labeled data is scarce.

  3. 2022

    Lexi for user interfaces

    In 2022, she coauthors Lexi, a pre-trained vision-and-language model designed for user-interface screens. Its UICaption dataset pairs 114,000 UI images with descriptions of their functionality.

  4. November 2023

    Teaching small models to reason

    In November 2023, she coauthors Orca 2, which teaches small language models to choose effective reasoning strategies for different tasks. She works as a Senior Research Software Engineer at Microsoft Research.

  5. 2024

    Contrastive post-training

    In 2024, she coauthors a NAACL Findings paper that constructs contrastive language-model training pairs from preferences across models of different strengths. The paper uses a curriculum that moves from easier to harder pairs to improve alignment.

  6. July–November 2024

    AgentInstruct and synthetic data

    In July 2024, she coauthors AgentInstruct, a framework that generates synthetic prompts and responses from raw text and code. By November 2024, Microsoft makes a one-million-pair subset of the 25-million-pair dataset publicly available.

  7. September 2026

    AI Night-Scientist

    In September 2026, she coauthors AI Night-Scientist, an agentic framework that uses reinforcement learning to teach models when and how to depart from predictable reasoning. It expands the range of research directions by 27.8% over the base model.

  8. As of October 2026

    Senior Research Software Engineer

    As of October 2026, she is a Senior Research Software Engineer in Microsoft Research's Redmond Data Science group. Her MSR projects include spiking and graph neural networks, multimodal language models, generative AI and large language models. Before joining MSR, she earns a master's in computer science from the University of California, Riverside.

Career

  1. NowMicrosoft ResearchSenior Research Software Engineer
  2. 2022–2026Microsoft

Education

  1. University of California, RiversideMaster's degree

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

Most-cited works

  1. Orca 2: Teaching Small Language Models How to Reason2023 · 30 citationsWith Ahmed Awadallah, Anastasia Razdaibiedina, Andres Codas, Corby Rosset, Erik Jones and Hamed Khanpour
  2. AgentInstruct: Toward Generative Teaching with Agentic Flows2024 · 4 citationsWith Ahmed Awadallah, Andres Codas, Corby Rosset, Dany Rouhana, Hamed Khanpour and Wei-ge Chen
  3. Lexi: Self-Supervised Learning of the UI Language2022 · 3 citations
  4. Learning without gradient descent encoded by the dynamics of a neurobiological model2021 · 2 citations
  5. Automatic Pair Construction for Contrastive Post-training2024 · Findings of the Association for Computational Linguistics: NAACL · 0 citationsWith Ahmed Awadallah and Corby Rosset

From OpenAlex, as of 21 October 2026.

  1. 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 Microsoft. 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.
  2. Google had the most people shaping AI every year from 2014 to 2025Includes Microsoft. 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.
  3. The company share of AI researchers’ top papers was 31% in 2020–2024 and 8% in 1970–1974Includes Microsoft. We estimate where the people shaping AI worked when their most-cited papers appeared: the company share was 31% in 2020–2024 and 8% in 1970–1974, and Google led all organizations in 2020–2024.
  4. Google has the highest combined h-index among AI organizationsIncludes Microsoft. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI organizations.

Colleagues

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

Sources and credits

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

  1. Learning without gradient descent encoded by the dynamics of a neurobiological model arxiv.org. Checked 21 October 2026.
  2. Leveraging semantically similar queries for ranking via combining representations arxiv.org. Checked 21 October 2026.
  3. Lexi: Self-Supervised Learning of the UI Language aclanthology.org. Checked 21 October 2026.
  4. Orca 2: Teaching Small Language Models How to Reason arxiv.org. Checked 21 October 2026.
  5. Orca 2: Teaching Small Language Models How to Reason - Microsoft Research microsoft.com. Checked 21 October 2026.
  6. Automatic Pair Construction for Contrastive Post-training arxiv.org. Checked 21 October 2026.
  7. AgentInstruct: Toward Generative Teaching with Agentic Flows arxiv.org. Checked 21 October 2026.
  8. Orca-AgentInstruct: Agentic flows can be effective synthetic-data generators - Microsoft Research microsoft.com. Checked 21 October 2026.
  9. Reinforcing Agentic Creativity in Scientific Ideation with Night Science arxiv.org. Checked 21 October 2026.
  10. Shweti Mahajan at Microsoft Research microsoft.com. Checked 21 October 2026.