Easey (Yizirui) Fang
Easey (Yizirui) Fang Easir Rey Fang

(they/them)
Amazon SDE and Applied ML Researcher
LLM Agents | Embodied AI | Uncertainty

About Me

I am an Amazon SDE and applied ML researcher focused on reliable AI agents, embodied/human-centered AI, and uncertainty-aware decision systems. I am interested in work that combines production-grade agentic systems, agentic RL, rigorous model evaluation, and trustworthy agents.

My strongest through-line is turning ambiguous model behavior into measurable scientific problems: code-generation agents and tool-use workflows in production settings, spoken instruction following for embodied agents, learning-to-defer systems for human-AI collaboration, and conformal prediction under data and distribution shifts.

Download CV
Interests
  • LLM Agents and Tool Use
  • Post-training and Evaluation
  • Human-centered Machine Learning
  • Embodied AI
  • Uncertainty Quantification
  • Robust Decision Making
Education
  • MS Eng in Computer Science

    Johns Hopkins University

  • Applied Statistics and Machine Learning

    Imperial College London

  • BSc Computer Science

    University of Nottingham

Selected Machine Learning Projects

Technical projects are written for fast hiring-manager review: problem, method, evaluation signal, and publication or artifact links where available.

Publication

Research outputs spanning uncertainty-aware decision making, embodied AI, human-centered AI, and conformal prediction.

(2026). Learning to Defer with an Uncertain Rejector via Conformal Prediction. Transactions on Machine Learning Research (TMLR).
(2026). Pragmatic Embodied Spoken Instruction Following in Human-Robot Collaboration with Theory of Mind. IEEE International Conference on Robotics and Automation (ICRA 2026).
(2025). ARCHED: A Human-Centered Framework for Transparent, Responsible, and Collaborative AI-Assisted Instructional Design. iRAISE 2025: Innovation and Responsibility for AI-Supported Education at the 39th AAAI Annual Conference on AI.
(2024). Investigating Data Usage for Inductive Conformal Predictors. arXiv preprint arXiv:2406.12262.