About

Hi, I’m Jen Wei.

I’m an AI research engineer and applied mathematician working across model architecture, optimization, memory, reinforcement learning, and distributed training. I approach research from both directions: asking why a system behaves as it does, then building the infrastructure and experiments required to find out.

I earned a PhD in Applied Mathematics and an MS in Statistics through a dual-degree program. That training taught me to find the underlying structure of a problem, expose the assumptions doing the real work, and distinguish genuine mechanisms from convenient explanations.

My work combines the depth of a scientist with the execution of an engineer. I formulate hypotheses, design the systems and experiments needed to test them, instrument training behavior, diagnose failures, and revise the research direction when the evidence disagrees.

Over the past decade, I have worked across three environments:

• Independent AI research, developed a Muon-based optimizer for GRPO-style reinforcement learning and extended olmo-core with Engram conditional memory to investigate architecture-specific effects on optimization, loss, and gradient stability. ››
• Startup R&D, joined a 3D vision SDK team as its first hire and took the product from zero to alpha release. ››
• Enterprise quantitative analysis, led quantitative modeling for energy and utilities while translating among regulators, traders, domain experts, and algorithms.

I am a speaker at PyTorch Conference North America 2026, presenting on Muon, Dion, and Orthogonalized Optimizers, and at PyData Global 2025, where I presented “I Built a Transformer from Scratch So You Don’t Have To.” My technical writing and teaching focus on making complex systems understandable without flattening the underlying ideas.

I’m most interested in research that demands both conceptual depth and hands-on execution: problems where the question is not yet clean, the system does not yet exist, and the answer has to be earned through careful experimentation.

You can find me on: