Hi, I’m Kuba Perlin. 👋
I’m currently working on reimagining the drug discovery process with AI at Isomorphic Labs, as a Senior Research Engineer.
I moved there in 2025 from the AlphaFold team at DeepMind.
Work Experience
During my 2+ years at Google DeepMind (2022-2025), I mostly worked on AlphaFold, contributing to the AlphaFold 3 paper, the AlphaFold Server and AlphaFold 3 Open Source.
Before joining DeepMind, I worked full time at Cohere, a start-up training and serving Large Language Models. Prior to that, I have interned at EPFL, Google, and NASA, among other places.
At NASA, I worked on applications of deep learning to computational fluid dynamics.
I have also worked as a Teaching Assistant (a.k.a. supervisor) at the University of Cambridge, teaching Probability and Complexity Theory courses in 2020, and Discrete Mathematics in 2021.
Last but not least, for 2 years, I was an after-school-class maths teacher preparing a cohort of gifted Polish students for the national Junior Math Olympiad. In my second year of teaching, my school performed, arguably, best in the country.
Co-authored Publications
- Accurate structure prediction of biomolecular interactions with AlphaFold 3 (2024, Nature)
J. Abramson et al. - Discovering Symbolic Cognitive Models from Human and Animal Behavior (2025, bioRxiv)
PS Castro et al. - Interlocking Backpropagation: Improving depthwise model-parallelism (2022, JMLR)
Aidan N. Gomez*, Oscar Key*, Kuba Perlin, Stephen Gou, Nick Frosst, Jeff Dean, Yarin Gal
- Scalable Training of Language Models using JAX pjit and TPUv4 (2022, arXiv)
Joanna Yoo*, Kuba Perlin*, Siddhartha Rao Kamalakara, João G.M. Araújo
Education
I studied Computer Science at the University of Cambridge (BA, Double First) and University of Oxford (MSc, Distinction).
My studies focussed on machine learning and theoretical computer science, including probabilistic algorithms, complexity theory, game theory, and computational learning theory.
I wrote my MSc thesis at the Oxford Robotics Institute, on 3D rotation invariance in neural networks for point cloud processing.
