PeopleofSuperintelligence

Researcher

Katerina Margatina

Also known as Aikaterini Margatina

Katerina Margatina is an NLP researcher whose work spans active learning for data-efficient language models, alignment data and LLM agents. Born and raised in Athens, she studied electrical and computer engineering at the National Technical University of Athens and completed a PhD at the University of Sheffield under Nikos Aletras, defending her thesis on active learning in 2024. Her EMNLP 2021 paper introduced Contrastive Active Learning, and the PRISM Alignment Dataset she co-authored won the Datasets and Benchmarks Best Paper award at NeurIPS 2024. After joining AWS Bedrock as an Applied Scientist in January 2024, she co-authored the CONFETTI benchmark for conversational function calling at ACL 2025. She is an Applied Scientist at Amazon in New York, where she works with the AWS AI Fundamental Research team on agentic AI, including AWS Security Agent.

Now at
Amazon
Badges

Timeline

  1. By 2019

    Engineering studies in Athens

    She grows up in Athens and studies Electrical and Computer Engineering at the National Technical University of Athens. While writing her diploma thesis on machine learning and natural language processing, she publishes a paper at ACL 2019 in Florence, "Attention-based Conditioning Methods for External Knowledge Integration", with Christos Baziotis and Alexandros Potamianos. Before her doctoral studies, she works as a Machine Learning Engineer at DeepSea Technologies.

  2. 2019–2024

    PhD at Sheffield

    In September 2019, she starts a PhD in the Natural Language Processing research group of the University of Sheffield's computer science department, supervised by Nikos Aletras and funded by an Amazon Alexa scholarship. Her topic is data-efficient NLP with active learning. Along the way she interns as a research scientist at Meta AI (FAIR) in London and at AWS in New York, and visits the CoAStaL group at the University of Copenhagen.

  3. 2021

    The CAL acquisition function

    Her EMNLP 2021 paper "Active Learning by Acquiring Contrastive Examples", written with Giorgos Vernikos, Loïc Barrault and Nikolaos Aletras, proposes Contrastive Active Learning (CAL). CAL selects unlabeled examples that sit close together in the model's feature space yet receive maximally different predicted likelihoods. It performs better than or equal to the best baseline across four natural language understanding tasks and seven datasets.

  4. January 2024

    Applied Scientist at AWS Bedrock

    She joins AWS Bedrock as an Applied Scientist in January 2024, working on LLM agents. Her focus is making agents more useful, reliable and efficient.

  5. July 2024

    A thesis on active learning

    She defends her PhD in July 2024 and passes with no corrections. Her thesis, "Exploring Active Learning Algorithms for Data Efficient Language Models", is built on four publications covering model training, data selection, in-context learning and the simulation of active learning, and it goes online in December 2024.

  6. 2024

    A NeurIPS best paper award

    The PRISM Alignment Dataset, co-authored with Hannah Rose Kirk, Paul Röttger and others, wins the Datasets and Benchmarks Best Paper award at NeurIPS 2024. PRISM maps the sociodemographics and stated preferences of 1,500 participants from 75 countries to their feedback in 8,011 live conversations with 21 LLMs, with census-representative samples for the UK and the US.

  7. 2025

    The CONFETTI benchmark

    CONFETTI, a benchmark for conversational function calling, is accepted at the ACL 2025 main track, with Tamer Alkhouli, James Gung and other co-authors. Its 109 human-simulated conversations cover 86 APIs and test the function-calling ability and response quality of large language models.

  8. As of October 2026

    AWS Security Agent

    She is an Applied Scientist at Amazon in New York, working with the AWS AI Fundamental Research team on agentic AI. On AWS Security Agent, which launches in preview at AWS re:Invent in December 2025, her work targets automated penetration testing of web apps and reliable vulnerability findings. In 2026, an ACL paper with Suhaib Abdurahman, Etsuko Ishii and others proposes Explicit Trait Inference to improve coordination among LLM-based agents.

Career

  1. Since 2024AmazonResearch Scientist
  2. DeepSea TechnologiesMachine Learning Engineer
  3. MetaResearch Scientist

Education

  1. 2024University of SheffieldPhD
  2. National Technical University of Athens

From the biography.

  1. 16% of AI career moves in the 2020s went from a university to a companyIncludes Katerina Margatina. 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.
  2. 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 Amazon. 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.
  3. 20% of people at companies who shape AI also hold a university role, led by Google and Stanford UniversityIncludes Amazon. 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.
  4. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Amazon. 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.

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 Katerina Margatina yet. Send us a free-licensed photo.

  1. My experience at Sheffield has definitely helped with my career | Computer Science | The University of Sheffield sheffield.ac.uk. Checked 21 October 2026.
  2. Aikaterini Margatina | ACM-W supporting, celebrating, and advocating for Women in Computing women.acm.org. Checked 21 October 2026.
  3. Attention-based Conditioning Methods for External Knowledge Integration - ACL Anthology aclanthology.org. Checked 21 October 2026.
  4. katerina margatina katerinamargatina.github.io. Checked 21 October 2026.
  5. [2109.03764] Active Learning by Acquiring Contrastive Examples arxiv.org. Checked 21 October 2026.
  6. Active Learning by Acquiring Contrastive Examples - ACL Anthology aclanthology.org. Checked 21 October 2026.
  7. Exploring Active Learning Algorithms for Data Efficient Language Models - White Rose eTheses Online etheses.whiterose.ac.uk. Checked 21 October 2026.
  8. NeurIPS 2024 Awards neurips.cc. Checked 21 October 2026.
  9. [2404.16019] The PRISM Alignment Dataset: What Participatory, Representative and Individualised Human Feedback Reveals About the Subjective and Multicultural Alignment of Large Language Models arxiv.org. Checked 21 October 2026.
  10. CONFETTI: Conversational function-calling evaluation through turn-level interactions - Amazon Science amazon.science. Checked 21 October 2026.
  11. Katerina Margatina - Amazon Science amazon.science. Checked 21 October 2026.