STEM Coding Expert (Physics/Math/Bio) at Mercor

STEM Coding Expert | $70/hr | Worldwide Remote

Mercor is partnering with a leading AI research organization to build a next-generation scientific coding benchmark. We're looking for advanced STEM domain experts to craft rigorous, research-derived computational challenges that evaluate frontier AI models on complex scientific reasoning tasks.

Key Responsibilities

• Author research-level scientific coding prompts derived from recent peer-reviewed work (post–July 2025)
• Decompose each problem into 3–5 sequential subproblems that reflect realistic scientific reasoning workflows
• Develop clean, well-documented, fully executable Python reference solutions with deterministic outputs
• Calibrate problem difficulty by evaluating model outputs against canonical solutions and iterating as needed
• Design comprehensive unit test suites ensuring 100% pass rates on reference solutions
• Tag problems with structured metadata including domain, subdomain, difficulty rating, and reviewer notes
• Participate in peer review to validate correctness, ambiguity resistance, and robustness against shortcut solutions
• Document structured multi-step solution trajectories demonstrating tool use and debugging behavior

Ideal Qualifications

• Advanced expertise in Physics, Chemistry, Mathematics, Biology, or a closely related technical field
• Strong ability to translate recent research into implementable computational challenges
• High proficiency in Python for scientific computing — simulations, numerical methods, data pipelines, symbolic computation
• Experience designing reproducible evaluation harnesses and robust unit tests
• Familiarity with dependency management and sandbox-safe execution environments
• Exceptional attention to detail with the ability to create unambiguous, deterministic problem specifications
• Ability to anticipate and mitigate shortcut solutions or memorization risks in AI evaluation contexts

This is a project-based contractor opportunity focused on producing high-difficulty, high-integrity evaluation data for cutting-edge AI systems.