William Li

Researcher @ OpenAI | CS @ Stanford | Nvidia DL Research | Jane Street Quant Research | Roo Code Eng

San Francisco Bay Area

About

I study CS (BS/MS) at Stanford with a sprinkle of math and English. My technical interests lie in deep learning and distributed computing, and I’ve done a lot of related research engineering work. At NVIDIA, I developed generative world models with multi-camera fusion to simulate tail risk, advancing the robustness of our AV fleet. At Jane Street, I used deep learning to find interpretable signals in new forms of text and options data. My research with Stanford’s Integrated Mental Health Lab was recently accepted to Nature Neuroscience, and my software engineering at various Series A startups (Beacons, TeachFX) has launched new features 0-1 to millions of users. I have a broad passion for socially impactful tech ranging from education (TeachFX) to mental health (Aprendi, IMHL) to housing justice (AEMP) and civic work (Interpreta, STV). Looking for ways to nurture human solutions to social problems with technology. Outside of work, I find inspiration in solarpunk, creative nonfiction, Mandopop, industrial sociology, topology (kinda), and Spider-Man!

Experience

  • Researcher at OpenAI
    Sep 2025 - Present · 11 mos

    Multi-Modal + Search Stuff

  • Member of Technical Staff at Roo Code
    Jun 2025 - Aug 2025 · 3 mos

    Building AI agents for code!

  • Language Model Researcher at Stanford Artificial Intelligence Laboratory (SAIL)
    Mar 2024 - Jun 2025 · 1 yr 4 mos

    Social and Language Technologies (SALT) Lab: Language Agents for Social Training Stanford Trustworthy AI Research (STAIR) Lab: LLMs for Low-Resource Languages

  • Deep Learning Research Scientist Intern at NVIDIA
    Aug 2024 - Dec 2024 · 5 mos

    Autonomous driving, World-Model Team: Building multi-camera sensor fusion for generative world model

  • Quantitative Research Intern at Jane Street
    May 2024 - Aug 2024 · 4 mos

    Trained LLM to extract relevant stocks from news. Built interpretable deep autoencoder for options trading.