Václav Volhejn

ML engineer/researcher. From papers to production. Various creative coding projects on the side. Personal website: https://vvolhejn.com/ YouTube: @PolylogCS (long-form CS education vids, >70k subscribers.) X/Twitter: @vvolhejn I'm a retired competitive programmer: I won a gold medal at the 2016 International Olympiad in Informatics, was an ICPC finalist in 2018 and 2020, and placed second at CodeVita 2021, the largest programming competition in the world. I designed the viral prompt injection game Gandalf (https://gandalf.lakera.ai/) and was part of the core team that worked on the game initially. Gandalf answered over 10M player messages within the first few months, and over 500k people have played the game. With a few friends we run the educational YouTube channel Polylog (https://www.youtube.com/@PolylogCS).

Notes4

• Quant employment verification: CONFIRMED. Václav Volhejn held a Software Engineering Internship at Jane Street in London in 2019 (LinkedIn experience entry), his only role at a quantitative-finance firm. Jane Street is a quantitative trading firm and electronic market maker (janestreet.com), so this is an actual quant-finance internship rather than a student insight day or camp. Identity is corroborated: the same LinkedIn profile links vvolhejn.com, whose About page carries the matching IOI 2016 Czech-Republic gold-medal biography and the same employer list. No other employer on his timeline (Kyutai, Lakera, impromptu.fun, Bini GmbH, ISTA, Datamole, Blue Vision Labs) operates in quantitative finance, banking, asset management, or energy/crypto trading.2019
• Joined Kyutai, the Paris non-profit open-science AI lab focused on audio research, as Technical Staff around March 2025, moving from Zurich; his work there includes the open-source Unmute voice-AI project.
• ML Scientist then Senior ML Scientist at Lakera (Zurich) 2022-2024, working on LLM security and the Gandalf prompt-injection game.
• Co-runs the Polylog educational YouTube channel (long-form CS videos), which had grown past 70k subscribers by 2025.
• Education: MSc in Computer Science at ETH Zurich (2020-2022), MSc thesis handed in summer 2022; BSc in Computer Science at Charles University in Prague (2017-2020).
• Interned at Blue Vision Labs (London) in 2017, before university; the company was later acquired by Lyft and folded into its Level 5 self-driving division.
• Scientific internship at ISTA (Klosterneuburg, Austria) in 2020 with prof. Christoph Lampert, producing the GCPR 2020 paper 'Does SGD Implicitly Optimize for Smoothness?'
• Maintains a personal site (vvolhejn.com) with dated technical/creative-coding write-ups including Unmute, MIRA world model, ColosSEAum, and a speech model 'Samuel'.Aug 17, 2026
• Began programming around age 8 with the Czech language Baltik, learned Java at ~12 via a kids' course, and took up competitive programming at ~14 through a Czech correspondence seminar.Feb 20, 2024
• Data Scientist Intern at Datamole (Prague) 2018-2019 and Research Engineering Intern at Bini GmbH (Berlin, computer-vision) in 2021.
• Led the team behind Impromptu (impromptu.fun), an AI image-guessing party game, in 2022-2023, overlapping his early Lakera period.
• Public profiles across GitHub (@vvolhejn), X (@vvolhejn), TikTok (@vvolhejn), and YouTube (Polylog + personal); the site is the hub linking all of them.