Armin Thomas
Co-founder & CTO at Radical Numerics
United States
Summary
🤩
Rockstar🎓
Top SchoolArmin Thomas is a machine learning founder and researcher with eight years of experience building AI systems end-to-end—from training large genomic and long‑context foundation models to shipping ultra‑efficient on‑device LLMs. As Co‑Founder of Radical Numerics and formerly Senior ML Scientist at Liquid AI, he led automated architecture search (STAR) and delivered Hyena Edge, a convolutional multi‑hybrid LLM optimized for edge latency and memory. His background spans top research labs (Stanford, Caltech, Max Planck) where he developed neuroscience- and biology-focused foundation models (e.g., Evo), long‑sequence convolutional alternatives to attention, and methods that integrate insights from neuronal plasticity into deep networks. Comfortable bridging rigorous research and product engineering, he combines a DSc in AI with practical experience training models at scale and shipping resource-constrained inference—often swapping assumptions common in NLP for biologically inspired, convolutional approaches.
9 years of coding experience
7 years of employment as a software developer
Master of Science (MSc), Cognitive Neuroscience, Master of Science (MSc), Cognitive Neuroscience at Freie Universität BerlinDoctor of Science (DSc), Artificial Intelligence and Machine Learning, Doctor of Science (DSc), Artificial Intelligence and Machine Learning at Technische Universität Berlin
English, German
Github Skills (20)
benchmarking10
self-supervised-learning10
neuroimaging10
transfer-learning10
natural-language-processing9
jupyter-notebook9
deep-learning8
post-training7
nvidia6
python6
pytorch6
nlp5
gpu5
kernel5
machine-learning5
Programming languages (2)
Github contributions (5)
athms/many-item-choice
Nov 2020 - Apr 2021
Uncovering the computational mechanisms underlying many-alternative choice
Contributions:96 commits, 2 PRs, 44 pushes in 5 months
athms/learning-from-brains
May 2022 - Jan 2023
Self-supervised learning techniques for neuroimaging data inspired by prominent learning frameworks in natural language processing + One of the broadest neuroimaging datasets used for pre-training to date.
Contributions:60 commits, 13 pushes, 5 comments in 7 months
natural-language-processingneuroimagingself-supervised-learningtransfer-learning
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.