Mathematician with Python Proficiency - AI Trainer

About the role

We're building a talent pool of Mathematics professionals with Python proficiency to contribute to project-based AI development initiatives, focused on evaluating and enhancing frontier AI models. Designed for mathematicians who enjoy deep technical problem-solving, this role involves applying analytical expertise to evaluate and push the boundaries of frontier AI models using domain-specific tools such as Z3, cvc5, SageMath, or Macaulay2, with verifiable, code-graded answers run inside isolated Linux environments.

Key Responsibilities

• Identify an appropriate mathematical software package and build problems whose solution genuinely hinges on that tool's core capabilities.
• Develop full Python solutions for each problem, providing input files where applicable.
• Establish the correct answer and define the acceptable margin for the AI model's response, based on the mathematical context.
• Run the problem against the AI model across multiple parallel attempts, analyzing where it succeeds or falls short, and adjust difficulty until the pass rate falls between 10% and 30%.
• Rewrite and refine problem parameters iteratively, building an understanding of how the model navigates complex mathematical challenges.
• Hand off completed tasks to a senior reviewer in your subfield and refine based on their feedback before final submission.

Core Requirements

• Academic background in Mathematics, Pure or Applied, or an equivalent field.
• At least 2 years of hands-on experience in mathematics research, applied work, or teaching.
• Solid Python skills, applied to writing and validating computational solutions.
• Capacity to build problems that cannot be solved without specialized mathematical software.
• Excellent written and verbal communication skills in English.
• Able to work independently in a remote, fast-paced environment.

Nice-to-Have

• Working knowledge of one or more domain-specific mathematical tools, including but not limited to Z3, cvc5, Macaulay2, Singular, CasADi, IPOPT, SDPB, G6K, fpylll, or SageMath, or a demonstrated ability to get up to speed independently.
• Prior exposure to how frontier AI models approach complex scientific problems.
• Knowledge spanning more than one area of mathematics, such as algebra, optimization, formal verification, or computational number theory.
• Familiarity with containerized or sandboxed Linux execution environments.