PeopleofSuperintelligence

Researcher

Juan Ciro

Also known as Juan Manuel Ciro

Juan Ciro is an AI and machine-learning engineer and researcher whose work spans speech datasets, evaluation benchmarks, language-model alignment, and agent frameworks. After building computer-vision tools for construction from 2017 to 2019 and manufacturing analytics at Iluma Alliance from 2019 to 2021, he served as tech lead for the Dynabench evaluation platform at MLCommons and Factored AI from January 2021 to May 2024. He coauthored the 50-language Multilingual Spoken Words Corpus and The People's Speech in 2021, DataPerf in 2022, and the 77-language Speech Wikimedia dataset and BabyLM Challenge findings in 2023. In 2024, he coauthored the Adversarial Nibbler red-teaming challenge and the PRISM Alignment Dataset, which won the NeurIPS 2024 Datasets and Benchmarks Best Paper award. Since May 2024, he is a technical lead and software and AI engineer at Contextual AI, where he leads the Agent Composer team, and in February 2026 he coauthored a Nature Medicine study testing LLM medical assistance with 1,298 participants.

Now at
Contextual AI
Badges

Timeline

  1. January 2017–January 2019

    Computer vision for construction

    From January 2017 to January 2019, he works as an ML and software engineer at Quantum Leap Data/Softytec, delivering predictive analytics and computer-vision solutions for construction management.

  2. January 2019–January 2021

    Manufacturing analytics at Iluma Alliance

    From January 2019 to January 2021, he works as an ML and software engineer at Iluma Alliance, building end-to-end analytics platforms and machine-learning solutions for manufacturing optimization.

  3. January 2021–May 2024

    Engineering Dynabench at MLCommons

    From January 2021 to May 2024, Ciro works as a senior software and ML engineer at MLCommons/Factored AI and serves as tech lead for the Dynabench AI evaluation platform. In May 2024, he helps rebuild base elements of Dynabench to handle many users concurrently for the PRISM project.

  4. 2021

    Multilingual and English speech datasets

    In 2021, he coauthors the Multilingual Spoken Words Corpus, an audio dataset of spoken words in 50 languages for keyword spotting and spoken-term search. In November 2021, he coauthors The People's Speech, a 30,000-hour conversational English speech recognition dataset for academic and commercial use. In December 2021, he is first author of a paper on locality-sensitive hashing models for deduplicating text data in an artificial Wikipedia dataset.

  5. 2022–2023

    DataPerf, Speech Wikimedia, and BabyLM

    In July 2022, Ciro coauthors the DataPerf benchmark suite for evaluating machine-learning datasets and data-centric algorithms, and by May 2023 he holds an engineering degree and a master's degree in artificial intelligence focused on deep learning from the International University of Applied Science. In August 2023, he coauthors Speech Wikimedia, a 77-language dataset of 1,780 hours of transcribed speech. In December 2023, he coauthors the BabyLM Challenge findings on sample-efficient language-model pretraining across more than 30 fixed-budget submissions, posted to arXiv in April 2025.

  6. 2024

    The PRISM Alignment Dataset

    In 2024, he coauthors the PRISM Alignment Dataset, which maps stated preferences and detailed feedback from 1,500 participants in 75 countries across 8,011 live conversations with 21 LLMs. The paper wins NeurIPS’s 2024 Datasets and Benchmarks Best Paper award.

  7. 2024

    Adversarial Nibbler and translation bias

    In 2024, Ciro coauthors Adversarial Nibbler, a red-teaming challenge that crowdsources implicitly adversarial prompts to identify harms in text-to-image generation. In August 2024, he coauthors the overview of a shared task evaluating gender bias in machine translation with multilingual holistic bias tests.

  8. Since May 2024

    Agent Composer at Contextual AI

    Since May 2024, he is a technical lead and software and AI engineer at Contextual AI, where he leads the Agent Composer team in designing modular, graph-based agent frameworks that convert RAG pipelines into scalable distributed systems. In February 2026, he coauthors a randomized study in Nature Medicine that tests LLM assistance for identifying conditions and choosing actions across ten medical scenarios with 1,298 participants.

Career

  1. Since 2024Contextual AITechnical lead and software/AI engineer
  2. 2021–2024MLCommonsSenior software/ML engineer and tech lead
  3. 2021–2024Factored AISenior software/ML engineer and tech lead
  4. 2019–2021Iluma AllianceML/software engineer
  5. 2017–2019Quantum Leap Data
  6. 2017–2019Softytec

From the biography.

  1. Google had the most people shaping AI every year from 2014 to 2025Our estimates of which organizations employed the most people shaping AI each year from 1955 to 2025, tracing the shift from university labs to companies.
  2. The United States employs an estimated 48% of people shaping AIWe estimate where the people shaping AI work today, how universities, companies and research institutes split across countries, and how country shares have changed since the 1960s.
  3. 43% of living people shaping AI work outside their country of birthWe estimate that 43% of living people shaping AI work outside their country of birth (520 of 1,198 people with known birth and employment countries).
  4. More people shaping AI studied at Massachusetts Institute of Technology than at any other institutionWe 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 Juan Ciro yet. Send us a free-licensed photo.

  1. Juan Ciro — Staff AI/ML Engineer jciro.com. Checked 22 October 2026.
  2. Unveiling the PRISM Alignment Project - MLCommons mlcommons.org. Checked 22 October 2026.
  3. Multilingual Spoken Words Corpus datasets-benchmarks-proceedings.neurips.cc. Checked 22 October 2026.
  4. [2111.09344] The People's Speech: A Large-Scale Diverse English Speech Recognition Dataset for Commercial Usage arxiv.org. Checked 22 October 2026.
  5. [2112.11478] LSH methods for data deduplication in a Wikipedia artificial dataset arxiv.org. Checked 22 October 2026.
  6. [2207.10062] DataPerf: Benchmarks for Data-Centric AI Development arxiv.org. Checked 22 October 2026.
  7. Adversarial Nibbler: A Data-Centric Challenge for Improving the Safety of Text-to-Image Models arxiv.org. Checked 22 October 2026.
  8. [2308.15710] Speech Wikimedia: A 77 Language Multilingual Speech Dataset arxiv.org. Checked 22 October 2026.
  9. Findings of the BabyLM Challenge: Sample-Efficient Pretraining on Developmentally Plausible Corpora - ACL Anthology aclanthology.org. Checked 22 October 2026.
  10. [2504.08165] Findings of the BabyLM Challenge: Sample-Efficient Pretraining on Developmentally Plausible Corpora arxiv.org. Checked 22 October 2026.
  11. [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 22 October 2026.
  12. NeurIPS 2024 Awards neurips.cc. Checked 22 October 2026.
  13. Adversarial Nibbler: An Open Red-Teaming Method for Identifying Diverse Harms in Text-to-Image Generation research.google. Checked 22 October 2026.
  14. Overview of the Shared Task on Machine Translation Gender Bias Evaluation with Multilingual Holistic Bias - ACL Anthology aclanthology.org. Checked 22 October 2026.
  15. Reliability of LLMs as medical assistants for the general public: a randomized preregistered study | Nature Medicine nature.com. Checked 22 October 2026.