His academic journey began with a BSc in Bioinformatics at Charles University, Prague. After graduating top of class, he pursued an MSc in Data Science at ETH Zurich. Considering this trajectory—from bioinformatics to data science to technical AI governance—one might suspect Evžen has developed a lifelong aversion to well-defined fields. When asked to confirm this theory, he offered no comment.
During his studies, Evžen twice took part in the MATS research program in Berkeley, working under researchers from Google DeepMind. He first led the creation of a dataset for evaluating vision-language models with David Lindner, and then, under Alex Turner, helped co-invent gradient routing—a method to localize knowledge in neural networks. This technical work sparked his interest in AI governance: gradient routing could implement the access controls he’d later argue could solve AI’s dual-use dilemma. He presented this position at the ICML 2025 TAIG Workshop. Through this work and the Talos AI Governance Fellowship, he’s found his niche at the intersection of technical AI safety and AI governance.
Looking ahead, Evžen aims to build tools that make the tradeoffs in AI governance more forgiving—for providers, users, and regulators alike. It’s a path that merges his love of hard technical problems with his knack for explaining them and connecting them to real-world applications, picking up on threads from high school debate trophies through teaching functional programming at his former high school to demystifying LLM architectures to students at ETH Zurich.
Beyond research, Evžen is passionate about photography, films, music, and books. He also still writes poetry from time to time, though he’s accepted he’s much better with Python than pentameter.