AI Benchmark Architect: Computational Mathematician

Mercor; San Francisco (CA); Remote; USD 83,000 - 165,000; Part time

Job summary

Mercor is seeking PhD and Master's scientists to author AI evaluation tasks (Sci Code) and contribute to a new benchmark for scientific computing in collaboration with leading AI labs. You will craft original, executable research problems that current frontier models struggle to solve.

You will source materials, write prompts, and define grading criteria, calibrating against models to ensure robust evaluation during a 6-week, part-time (20+ hours/week) engagement with immediate start.

Qualifications

• PhD in mathematics, applied mathematics, computational mathematics, or closely related field.
• Demonstrated depth in at least two of the following subdomains: numerical linear algebra, computational mechanics, computational finance.
• Working proficiency in Python or R for scientific computing.
• Comfortable with Git/GitHub and running code in Docker — authoring runs through a pull-request workflow with automated quality checks.

Responsibilities

• Source your own material: a published paper, a Kaggle dataset, an open-source repository, or a scenario you design.
• Write scientific prompts based on the input
• Build the grading criteria that define a correct answer
• Calibrate against frontier models — a task ships only when strong models fail it more often than they succeed

Skills

Numerical linear algebra

Computational mechanics

Computational finance

Python or R for scientific computing

Git

Education

PhD in mathematics, applied mathematics, computational mathematics or closely related field

Tools

Docker

GitHub

Job description

Mercor is seeking PhD and Master's scientists to author AI evaluation tasks (Sci Code) and contribute to a new benchmark for scientific computing in collaboration with leading AI labs. You will craft original, executable research problems that current frontier models struggle to solve.

You will source materials, write prompts, and define grading criteria, calibrating against models to ensure robust evaluation during a 6-week, part-time (20+ hours/week) engagement with immediate start.