Researcher

James Fan

James Fan is an AI researcher whose work spans question answering, game-playing systems, speech analysis and LLM-agent security. After earning a PhD at UT Austin, he coauthored IBM's DeepQA overview and Watson strategy research; his later work includes a self-play poker system and speech-sentiment research. The IBM Watson team received the 2013 AAAI Feigenbaum Prize, and Fan went on to work on AI-powered customer support and co-found Tomato.ai, whose voice filter softens offshore call-center agents' accents. As of October 2026, he researches how to make LLM agents safe to deploy.

Now at
Tomato.ai
Works
54
Citations
2,669
h-index
17

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

  1. 2009

    The Loose-Speak Interpreter

    In 2009, he coauthors “Automatic interpretation of loosely encoded input” with Ken Barker and Bruce Porter. Its Loose-Speak Interpreter reaches 98% average precision and 90% average recall in empirical studies.

  2. 2011

    Watson and DeepQA

    In 2011, he is a member of IBM's Watson team and holds a PhD from The University of Texas at Austin. He coauthors the DeepQA overview, which describes Watson as a real-time Jeopardy! contestant built to compete at the human-champion level.

  3. 2013

    A prize for IBM Watson

    The IBM Watson team receives the 2013 AAAI Feigenbaum Prize for showing that a synthesis of AI techniques can reach human-level real-time factual question answering; Fan is named among its members. He coauthors a 2013 study that analyzes Watson's Jeopardy! strategy through simulation, learning and Monte Carlo methods.

  4. 2016

    Poker-CNN learns through self-play

    At Columbia University, he coauthors Poker-CNN in 2016, a poker system that learns through iterative self-play and improves to compete against human experts.

  5. 2017

    HelloVERA in AI Nexus Lab

    In 2017, HelloVERA joins NYU and ffVC's AI Nexus Lab cohort as an AI-powered customer-support startup. Fan and LiangLiang Cao point to the lab's presentation guidance.

  6. 2020

    Speech sentiment analysis

    In 2020, he coauthors a Google-affiliated study on pre-trained end-to-end ASR features for speech-sentiment analysis and a Switchboard speech-sentiment corpus. The corpus adds three human annotators' sentiment labels to each transcript segment and contains 49,500 labeled segments across 140 hours of audio.

  7. March 2024

    Co-founding Tomato.ai

    In March 2024, he is Tomato.ai's co-founder and CTO. Its voice filter softens offshore call-center agents' accents as they speak, and Ofer Ronen credits Fan with the seed idea of using a French accent to sell wine.

  8. As of October 2026

    LLM agent security research

    In July 2026, his LangGraph Agent Security series examines software-dependency and supply-chain risks. As of October 2026, he researches how to make LLM agents safe to deploy, including risks that arise when agents can use real tools; his earlier roles include leading Google Cloud's Speech Group and teaching at Columbia University.

Career

  1. From 2024Tomato.aiCo-founder, CTO
  2. 2019–2020GoogleGroup Lead
  3. 2019–2022Naval Postgraduate School
  4. 2012–2017Pennsylvania State University
  5. 2011–2014IBMTeam Member
  6. Columbia UniversityTeacher

Education

  1. University of Texas at AustinPhD

Most-cited works

  1. Building Watson: An Overview of the DeepQA Project2010 · AI Magazine · 1,523 citationsWith Adam Lally, Aditya Kalyanpur, Chris Welty, David Ferrucci and David Gondek
  2. Efficient Clustering of Very Large Document Collections2001 · Massive computing · 240 citationsWith Inderjit S. Dhillon
  3. Fluorescence Excitation Emission Matrices of Human Tissue: A System for in vivo Measurement and Method of Data Analysis1999 · Applied Spectroscopy · 114 citations
  4. Project Halo: Towards a Digital Aristotle2004 · 84 citationsWith Michael Witbrock and Steffen Staab
  5. Speech Sentiment Analysis via Pre-Trained Features from End-to-End ASR Models2020 · IEEE International Conference on Acoustics Speech and Signal Processing · 79 citations

Sources and credits

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

  1. Automatic interpretation of loosely encoded input - IBM Research research.ibm.com. Checked 21 October 2026.
  2. Three cheers for Watson - UT News news.utexas.edu. Checked 21 October 2026.
  3. Building watson: An overview of the deepQA project for AI Magazine - IBM Research research.ibm.com. Checked 21 October 2026.
  4. AAAI Feigenbaum Prize - AAAI aaai.org. Checked 21 October 2026.
  5. Analysis of Watson’s strategies for playing Jeopardy! for JAIR - IBM Research research.ibm.com. Checked 21 October 2026.
  6. Poker-CNN: A Pattern Learning Strategy for Making Draws and Bets in Poker Games Using Convolutional Networks - Proceedings of the AAAI Conference on Artificial Intelligence ojs.aaai.org. Checked 21 October 2026.
  7. ffVC and NYU are mixing researchers with investors in the name of machine intelligence - TechCrunch techcrunch.com. Checked 21 October 2026.
  8. IEEE ICASSP 2020 || Barcelona, Spain || 4-8 May 2020 cmsworkshops.com. Checked 21 October 2026.
  9. A Large Scale Speech Sentiment Corpus - ACL Anthology aclanthology.org. Checked 21 October 2026.
  10. Ofer Ronen, Co-Founder and CEO of Tomato.ai – Interview Series – Unite.AI unite.ai. Checked 21 October 2026.
  11. James Fan — AI Agents, LLMs, GenAI & Security Research jamesfan.net. Checked 21 October 2026.
  12. 17. Dependency and Supply Chain Security: The Threat Hiding in Your Requirements File — James Fan jamesfan.net. Checked 21 October 2026.