PeopleofSuperintelligence

Builder

Jake Poznanski

Also known as Jakub Poznanski

Jake Poznanski is a software engineer and founder whose AI work turns PDFs into clean text for training open language models. Born in Poland, they studied computer science at Carnegie Mellon, co-founded the game studio Random Salad Games in 2011, and founded Pozotron, a machine-learning tool for proofing audiobooks, in 2017. After joining the Allen Institute for Artificial Intelligence (Ai2) as a researcher in 2024, they were first author of olmOCR in 2025, an open-source toolkit that uses a 7B vision language model to convert PDFs into plain text. Its successor, olmOCR 2, trained its model with unit-test rewards in 2025, and by 2026 the system had processed more than 100 million PDFs to curate pretraining data for Olmo 3, a family of fully open language models they coauthored.

Timeline

  1. 2007–2011

    Computer science at Carnegie Mellon

    Born in Poland, Poznanski lives in the United States for most of their life and studies computer science at Carnegie Mellon University from 2007 to 2011, graduating with a BSc in 2011. They intern at QinetiQ North America in 2009 and at Microsoft in 2010.

  2. 2011

    Random Salad Games and SpaceX

    In 2011, they co-found Random Salad Games with fellow Carnegie Mellon computer science graduate Sam Kaufmann and serve as its CTO. The studio starts with Windows Phone games and becomes one of the largest indie studios on the Microsoft Store. Poznanski also works as a flight software engineer at SpaceX in 2011.

  3. 2017–2019

    Founding Pozotron

    In 2017, Poznanski founds Pozotron to speed up quality checks for audiobooks and serves as its founder and chairman. By 2019, its proofing tool uses the company's own machine learning model to check an audio recording against its script, flagging missing words, added words and long pauses.

  4. 2024

    A researcher at Ai2

    In May 2024, they release an RNN-T speech recognition system that runs in the web browser with TensorFlow.js. Poznanski joins the Allen Institute for Artificial Intelligence (Ai2) as a researcher in 2024.

  5. February 2025

    The olmOCR toolkit

    Released in February 2025, olmOCR is an open-source toolkit that turns PDFs into clean plain text in natural reading order using a fine-tuned 7B vision language model, at a cost of about 176 US dollars per million pages. Poznanski is the first of its ten authors, and Ai2's AllenNLP team develops and maintains the toolkit.

  6. October 2025

    Unit-test rewards in olmOCR 2

    In October 2025, Poznanski, Luca Soldaini and Kyle Lo release olmOCR 2, powered by olmOCR-2-7B-1025, a 7B vision language model trained with reinforcement learning whose rewards are binary unit tests. The model reaches state-of-the-art results on the olmOCR-Bench benchmark, with the largest gains on math formulas, tables and multi-column layouts.

  7. December 2025

    Coauthoring Olmo 3

    In December 2025, they appear among the authors of Olmo 3, a family of fully open language models at 7B and 32B parameters. The release covers every stage, checkpoint and data point used to build the models, with Olmo 3 Think 32B as the flagship thinking model.

  8. July 2026

    An olmOCR paper at ACL

    In July 2026, Poznanski, Kyle Lo and Luca Soldaini publish “The olmOCR Project: Building Fully Open OCR using VLMs” in the system demonstrations volume of ACL 2026 in San Diego. The paper reports that olmOCR has run on more than 100 million PDFs to curate pretraining data for Olmo 3.

  1. Google had the most people shaping AI every year from 2014 to 2025Our estimates of which organizations employed the most people shaping AI from 1961 to 2025, with a year-by-year chart.
  2. The United States employs an estimated 53% of people in our data with a known employment countryOur estimates of where AI careers are concentrated now and how countries’ shares of dated positions changed over time.
  3. 46% of living people in our data with known birth and employment countries work outside their birth countryOur estimate of cross-border careers in AI, from country of birth and the country of each current organization.
  4. More people shaping AI studied at Massachusetts Institute of Technology than at any other institutionOur estimates of where the people shaping AI studied and where their careers led, from education records in cited biographies.

Sources and credits

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

  1. About | Jake Poznanski — Robots, Thoughts, AI Rights jakepoz.com. Checked 19 October 2026.
  2. Jake Poznanski - Founder at Pozotron | The Org theorg.com. Checked 19 October 2026.
  3. Random Salad Games newpage.randomsaladgames.com. Checked 19 October 2026.
  4. Meet The Pozotron Team | AI-Powered Audio Proofing Software pozotron.com. Checked 19 October 2026.
  5. Pozotron Audiobook Proofing | Jake Poznanski — Robots, Thoughts, AI Rights jakepoz.com. Checked 19 October 2026.
  6. RNN-T Speech Transcription in the Browser | Jake Poznanski — Robots, Thoughts, AI Rights jakepoz.com. Checked 19 October 2026.
  7. Jake Poznanski — Lacuna lacuna.tiptreesystems.com. Checked 19 October 2026.
  8. [2502.18443] olmOCR: Unlocking Trillions of Tokens in PDFs with Vision Language Models arxiv.org. Checked 19 October 2026.
  9. GitHub - allenai/olmocr: Toolkit for linearizing PDFs for LLM datasets/training github.com. Checked 19 October 2026.
  10. [2510.19817] olmOCR 2: Unit Test Rewards for Document OCR arxiv.org. Checked 19 October 2026.
  11. [2512.13961] Olmo 3 arxiv.org. Checked 19 October 2026.
  12. The olmOCR Project: Building Fully Open OCR using VLMs - ACL Anthology aclanthology.org. Checked 19 October 2026.