I’m an Agent Engineer at Sierra.ai. I graduated with a Master’s in Electrical Engineering and Computer Sciences from UC Berkeley in May 2025, advised by Professor David Wagner. My research focused on training language models to follow an instruction hierarchy, specifically looking at improving system prompt robustness in LLMs and prompt injection robustness in reasoning models. Before that, I completed my Bachelor’s degree in Computer Science, also at UC Berkeley.

Research

  1. A Closer Look at System Prompt Robustness
    Norman Mu*, Jonathan Lu*, Michael Lavery, David Wagner
    arXiv preprint; NeurIPS Safe Generative AI Workshop

Projects

  • Reward Modeling From GPT4-Vision Preferences
    | Writeup | GitHub |

Work Experience

  • Sierra.aiAgent Engineer
    San Francisco, CA • Jun 2025 - Present
  • Berkeley AI Research, Wagner LabResearch Assistant
    Berkeley, CA • May 2024 - Aug 2024
  • DeepgramSoftware Engineering Intern
    Ann Arbor, MI • May 2023 - Aug 2023
  • MetaEnterprise Engineering Intern
    Sunnyvale, CA • May 2022 - Aug 2022

Teaching

  • CS 189 Machine Learning — Spring 2023
  • CS 61A Structure and Interpretation of Computer Programs — Fall 2022