RESEARCHED WITH AUTUMN
Nikita Khomich
Scaling RL @Tzafon
Tzafon ·
United Kingdom · engineer
Founding Member Of Technical Staff At Tzafon, Working On Scaling Reinforcement LearningBronze Medal For Belarus At IMO 2020 (rank 262, 17 Points)University Of Cambridge (2021-2024)Co-author Of 'Grokking Revisited' (LSTM Reverse-engineering, Dec 2024)Speaker On Robust Computer-use Agents (Lisbon AI, Feb 2026)
Nikita Khomich (NikitosKh) — founding member of technical staff at Tzafon (AI, scaling RL); Cambridge mathematics graduate; IMO 2020 bronze medallist for Belarus; co-author of 'Grokking revisited' (LSTM reverse-engineering); speaker on robust computer-use agents.
Reinforcement learningMachine learning researchDeep learningMathematicsQuantitative researchComputer-use agentsAutomated theorem provingStatisticsMechanistic interpretability
- Quant employment verification: UNRESOLVED. No employer or role can be confirmed from primary sources. The only quantitative-finance-adjacent entries are LinkedIn-index-only: a 'Quantitative Research' role 2024-06..2025-01 at a company left Undisclosed (no employer name, no domain, no primary-source page), and a 'Spring Intern' at Citadel listed 2022 (Citadel is a genuine hedge fund/market maker, but a 'spring intern' is commonly a short insight/spring-week student programme, which does not by itself establish employment). Neither has independent primary-source corroboration; the LinkedIn enrichment index is not an independent confirmation. Retention: uncertainty retained.
- Founding Member of Technical Staff at Tzafon (AI, San Francisco Bay Area), stated to be scaling reinforcement learning; this is a current technical/engineering role, not a quantitative-finance role.
- Co-author of 'Grokking revisited: reverse engineering grokking modulo addition in an LSTM' (Dec 2024), a mechanistic-interpretability study with public code, indicating active ML research output.Dec 2024
- Bronze medal for Belarus at the 61st International Mathematical Olympiad (IMO 2020, held online), confirmed by the official IMO results page (rank 262, 17 points).Sep 2020
- Studied at the University of Cambridge 2021-2024 (mathematics), per the LinkedIn index; no degree title is stated in sources, so no degree is asserted.
- Prior research and engineering roles listed in the LinkedIn index: Junior Researcher at Huawei (Cambridge, automated theorem proving, 2022); Research Intern at IPAM, UCLA (DL for time series denoising, 2022); Computational Scientist contractor at University of Cambridge (applied statistics/ML, 2023). Index-only, no independent primary-source page read for these.
- Graduated from Lyceum BGU (Lyceum of Belarusian State University, Minsk); the Lyceum's own Facebook page announced his bronze medal as a graduate of the school.Sep 2020
- Public code presence at github.com/NikitosKh (username also used as an alias), consistent with the ML/engineering profile.
Experience
Founding Member Of Technical Staff at Tzafon
San Francisco Bay Area
Research / infra
Quantitative Research at Undisclosed
London Area, United Kingdom
Computational Scientist contractor at University of Cambridge
Cambridge, England, United Kingdom
Applied statistics/ml research
Research Intern at Institute for Pure and Applied Mathematics
Los Angeles County, California, United States
DL for time series denoising
Spring Intern at Citadel
Junior Researcher at Huawei
Cambridge, England, United Kingdom
Automated theorem proofing
Education
University of Cambridge
Lyceum of Belarusian State University (Lyceum BGU), Minsk
secondary / high school
Projects
AchievementsBronze medal, 61st International Mathematical Olympiad (IMO 2020), representing Belarus (rank 262, 17 points) · Graduate of Lyceum BGU (Minsk), Belarus · Co-author of 'Grokking revisited: reverse engineering grokking modulo addition in LSTM' (LessWrong, Dec 2024) · Speaker, 'The Hunt for Robust Computer Use Agents' (Lisbon AI, Feb 2026)
Articles
Talks
AwardsBronze medal, IMO 2020 (Belarus)
Profiles
Competition record
Belarus · IMO