AutumnTHE OLYMPIAD COLLECTION
Menu
RESEARCHED WITH AUTUMN
MZ

Miklós Zsigmond Horváth

Senior Research Engineer at Google DeepMind
Google DeepMind · London Area, United Kingdom · engineer

Miklós Z. Horváth is a Senior Research Engineer at Google DeepMind in London, working on reinforcement learning and large language models. He created the test-time RL (TTRL) inference method behind AlphaProof, which reached silver-medal-level performance at IMO 2024 (Nature 2025). His personal site lists contributions to Formal Conjectures, AlphaProof Nexus, DeepMind's IMO 2025 gold-medal effort, the Gemini 3 family, and SynthID-Image. He holds a BSc (2016-2019) and MSc (2019-2022) in Computer Science from ETH Zürich, and represented Austria at IMO 2015 (Honourable Mention) and IMO 2016 (Bronze, 17 pts).

Details
LocationLondon Area, United Kingdom
Company sitedeepmind.com
UniversityETH Zürich
Reinforcement LearningLarge Language ModelsMachine LearningMathematicsFormal Proof Search
Notes
  • Quant employment verification: CONFIRMED. LinkedIn/index profile lists a past role 'Quantitative Research Intern' at Jump Trading LLC (London, 2021), described as using machine learning for research projects with trading applications on large financial datasets; his personal site miklos.ai states he 'completed internships as a software developer at Amazon and a quantitative researcher at Jump Trading' during his Master's. Identity is supported by the matching ETH Zürich CS MSc/BSc timeline and IMO 2016 bronze. Employer classification: Jump Trading LLC is a proprietary algorithmic/high-frequency trading firm, an operating quantitative trading company (not merely a name containing 'quant'). Caveat: the role is documented via LinkedIn/index and his own site, not an independent third-party primary source; no exact internship start/end month is stated.2021
  • Mathematical Olympiad record (primary source, imo-official.org): represented Austria in 2016 (rank 224, 63.0%, Bronze B, 17 pts: P1=7,P2=3,P4=7) and 2015 (rank 365, 36.9%, Honourable Mention HM, 9 pts: P1=7,P4=2).2016
  • Personal site miklos.ai is the authoritative self-hosted bio: it states he created the test-time RL (TTRL) inference method behind AlphaProof and lists contributions to Formal Conjectures, AlphaProof Nexus, DeepMind's IMO 2025 gold-medal effort, the Gemini 3 family, and SynthID-Image; these are first-party self-reported claims, not independent confirmation.
  • Austrian OeMO records list him at BG/BORG Feldkirch, Schillerstraße (Vorarlberg): 2015 Bundeswettbewerb qualified him for the 56th IMO (2015) and 2016 Bundeswettbewerb qualified him for the 57th IMO (2016), documenting his school and the national selection path.2016

Competition record

Austria · IMO

Related collections

Explore the collection