About this role
Help shape the future of AI by designing challenging math problems that test and improve reasoning in cutting-edge models.
Location: Remote
Type: Contract / Part-time
Commitment: 20 hours per week
Compensation: Up to 40 USD / hr
Project duration: 2 months, with potential extension
Availability: Immediate start
About the roleWe create high-quality STEM training data for frontier AI models. Our data is used directly in training and evaluation pipelines at leading AI labs to improve model reasoning in technical domains.
We are looking for experts in Mathematics to design rigorous, deterministic problems that are genuinely challenging for state-of-the-art AI systems. Each problem must have exactly one verifiable correct answer and be submitted together with a complete, verified solution.
What you’ll doDesign advanced mathematical problems for frontier AI training and evaluation
Create deterministic problems with exactly one correct answer
Write complete, verified solutions and clearly document the reasoning process
Develop problems that test deep mathematical reasoning, not just memorization
Where relevant, use Python or specialized tools to build computational workflows
Ensure all outputs are technically precise, reproducible, and well-written in English
What we’re looking forMaster’s or PhD in Mathematics or a closely related field
Strong research or industry experience involving mathematical modeling, proof-based reasoning, applied mathematics, or computational mathematics
Strong Python skills; comfort with libraries such as numpy, scipy, pandas, or similar
Solid grasp of algorithms, numerical methods, and computational approaches
Ability to design original, difficult problems that mirror real mathematical workflows
Excellent attention to detail and technical writing skills in English
Nice to haveExperience with symbolic math systems, theorem provers, optimization solvers, or other mathematical software
Background in olympiad-level, graduate-level, or research-level problem design
Experience evaluating model reasoning, benchmarking, or technical assessment design