Computational Astrophysics Researcher for AI Benchmark Design

Role Overview

Design original, graduate-level computational astrophysics and cosmology problems that evaluate whether advanced AI systems can perform research-level scientific work with real software. You will create and refine tasks involving simulations, result interpretation, experiment design, and extracting hidden information from data.

Key Responsibilities

• Create challenging problems requiring skilled use of specialized scientific software.
• Develop fully defined tasks with exact answers that test complex, multi-step scientific workflows.
• Design open-ended investigation tasks in which an AI must plan queries or experiments, interpret partial results, choose useful measurements, and efficiently narrow possible answers.
• Test each problem against advanced AI models and refine its design until it reaches the intended level of difficulty.
• Write Python-based problem setups, oracle functions, and solution validators.

Qualifications

• Graduate-level training in a relevant STEM field, including an MS, PhD, or equivalent research experience. An MS or PhD is preferred.
• Deep hands-on experience in astrophysics and cosmology, including astropy and related tools for cosmological calculations, angular power spectra, galaxy survey analysis, and observational data-reduction pipelines. Experience with other specialized domain software will also be considered.
• Demonstrated proficiency with at least one relevant scientific software library through research publications, open-source contributions, or professional work.
• Strong Python skills and practical experience using scientific libraries to solve real research problems.
• Ability to recognize software limitations and edge cases, and to design problems that reward careful reasoning rather than surface-level pattern matching or raw computation.
• Ability to work independently, incorporate feedback, and refine problem designs.
• Comfort working in a Linux terminal environment with remote compute sandboxes.

Preferred Qualifications

• Experience across multiple relevant scientific domains or tools.
• Familiarity with benchmark or evaluation design.
• Experience teaching science or designing exams and problem sets.
• Experience with computational reproducibility and containerized environments.

Work Terms

• Remote, hourly engagement.
• Availability of at least 15 to 20 hours per week.

Compensation

$70 to $100 per hour.