Kevin Lu
PhD student · Northeastern University
NSF GRFP Fellow
I'm a PhD student at Northeastern working on AI for biology, advised by David Bau and Wengong Jin. I use mechanistic interpretability to understand protein and cell foundation models, and I want to use what we learn to accelerate drug discovery.
Most recently, I've been reverse-engineering AI protein folding models (ESMFold, OpenFold, Boltz-1) to find out how they turn sequence chemistry into 3D structure.
lu.kev@northeastern.edu
Scholar
GitHub
Twitter
LinkedIn
Steering charge inside ESMFold
α = 0.00σ
News
- Sep 2026NeurIPSTwo Stages of Folding: Convergent Mechanisms in AI Protein Folding Trunks was accepted to NeurIPS 2026! The new version extends our ESMFold analysis to OpenFold and Boltz-1.
- Feb 2026Presented our protein folding work at the Flatiron Institute.
- Sep 2025NeurIPSWhen Are Concepts Erased From Diffusion Models? was accepted to the NeurIPS 2025 main track.
- Jul 2025Started a 6-month internship at Takeda Pharmaceuticals on protein language model interpretability for antibody design.
Reviewing
- Main Conference (AI for Science)
- Scientific Methods for Deep Learning Workshop
- Main Conference (Geometric Deep Learning)
- Mechanistic Interpretability Workshop