Andrei Kanavalau

I’m a PhD candidate in Electrical Engineering at Stanford, in the Lall Group. My research sits at the intersection of machine learning, optimization, and control theory, driven by a fascination with complex systems and a desire to understand their behavior well enough to build safer and more efficient systems. I expect to graduate in June 2026 and am currently looking for full-time opportunities. Prior to my PhD, I earned a B.A. and M.Eng. in Chemical Engineering from the University of Cambridge in 2019, and an M.S. in Electrical Engineering from Stanford in 2023. Drawing on my interdisciplinary background, I’ve had the chance to apply the same core ideas across very different settings, from thermal stability of exothermic batch processes, to constrained-optimization methods for enforcing structure and constraints in deep learning, to more recently simplifying LLM architectures while preserving training stability. Along the way I’ve worked on applied problems at TSMC, Applied Materials, KLA, and Inflection AI.

Some things I have worked on recently

TaperNorm

Gated removal of normalization in Transformers for stable training and inference-time folding.

Parameter-efficient Adaptation of Tokenizer-free Byte Latent Transformer

Can a tokenizer-free architecture be adapted to new languages by retraining only the ~4% “interface” modules and keeping the core transformer fixed?

Safe Explicit Model Predictive Control via Constrained Neural Network Training

Fast MPC policies learned offline with safety constraints using a primal-dual (augmented Lagrangian) training loop.