What are the responsibilities and job description for the ML Engineer, Math Formalization & Reward Modeling position at Millennium Research?
Company Description
Millennium Research is dedicated to pushing the frontier of mathematics using advanced AI methods. The organization focuses on formalizing mathematical concepts in Lean 4 and mathlib and building the verification infrastructure needed to certify that formal statements faithfully capture the mathematics they derive from. Team members collaborate at the intersection of theoretical mathematics, formal verification, and AI to develop tools that expand what is computationally and mathematically possible. The environment is research-driven, with an emphasis on rigorous experimentation, reproducibility, and open scientific dialogue. Applicants can expect to work on innovative projects that have the potential to reshape how mathematical discovery is conducted.
Role Description
The Formalization Engineer, Math & Reward Modeling role is a contract, on-site position based in New York, NY. In this role, the engineer formalizes mathematical statements and proofs from research papers into Lean 4/mathlib, evaluates formal statements for correctness and faithfulness to source material, and helps build the datasets that underpin the company's reward modeling work. Day-to-day responsibilities include translating LaTeX proofs and theorem statements into verified Lean code, reviewing and adjudicating borderline or disputed formalizations, and maintaining consistency across a growing corpus of formal/informal pairs. The engineer will also support dataset curation and labeling for downstream reward model training, though this is a secondary responsibility; the bulk of the role is formalization and verification work. The role involves documenting methods, contributing to research reports, and participating in technical discussions that shape the research program's direction.
Qualifications
- Strong command of Lean 4 and mathlib, with a track record of formalizing nontrivial theorems or proofs.
- Solid background in pure mathematics (e.g., abstract algebra, analysis, logic, topology, combinatorics, or number theory) at the level of upper-division coursework or competition mathematics.
- Comfort reading and interpreting research-level mathematical papers, including translating informal proof sketches into rigorous formal statements.
- Familiarity with proof assistants and type theory fundamentals (dependent types, tactics, term-mode proofs).
- Working knowledge of Python for basic data handling and tooling; no deep ML background required.
- Meticulous attention to detail; able to identify subtle gaps between an informal statement's intent and its formalization.
- Ability to collaborate closely in an on-site research environment, communicating clearly with interdisciplinary teams.
- Advanced degree (or equivalent competition/research experience) in Mathematics, Computer Science, or a related field.
- Prior exposure to reinforcement learning, reward modeling, or ML dataset curation is a plus but not required.