Physical Sciences AI Benchmark Task Expert

The work

You will create realistic terminal-based scientific tasks for AI training and evaluation. The work turns real physics, chemistry, materials science, astronomy, and computational science workflows into reproducible benchmark environments.

You will combine scientific modeling, software development, automated grading, and expert review. You will assess whether AI agents can reason through problems, use command-line tools, debug calculations, and produce reliable scientific files.

• Design multi-step terminal tasks based on real physical-science workflows.
• Build self-contained computational environments with fixed dependencies and scientific software.
• Create datasets, molecular structures, simulation settings, experimental data, and model configurations.
• Write expert solutions using Python, Bash, C/C++, Julia, or domain-specific tools.
• Develop automated tests for numerical accuracy, physical consistency, convergence, and output structure.
• Create tasks involving simulations, numerical modeling, data fitting, optimization, spectroscopy, molecular analysis, and scientific visualization.
• Define numerical tolerances, units, boundary conditions, and expected scientific behavior.
• Validate reproducibility and debug dependency, precision, solver stability, performance, and file-format problems.

What it pays and takes

The listing does not specify a pay rate. This is a part-time contractor role with a 20+ hour weekly commitment, open worldwide.

• Pay: Not specified in the listing.
• Schedule: 20+ hours per week.
• Location: Worldwide.
• Language: English.
• Experience level: Listed as entry level, with advanced technical experience required for the work.
• Education or experience: A Ph.D., postdoctoral experience, or equivalent advanced technical experience in a relevant physical-science field.
• Programming: Strong ability in Python, C/C++, Julia, Bash, or another scientific programming language.
• Technical background: Linux or terminal-based environments, numerical methods, scientific modeling, simulations, or quantitative data analysis.
• You must be able to independently create and validate computational scientific workflows.
• You must understand units, numerical precision, physical constraints, and scientific reproducibility.

Helpful background

Experience with scientific Python libraries, simulation methods, optimization, or statistical modeling is useful. Familiarity with research software engineering, benchmark creation, automated graders, AI coding-agent evaluation, publications, or open-source computational science also supports this work.

• NumPy, SciPy, pandas, matplotlib, SymPy, or JAX.
• Molecular dynamics, quantum chemistry, finite-difference methods, Monte Carlo methods, optimization, or statistical modeling.
• OpenMM, ASE, RDKit, Psi4, LAMMPS, or GROMACS.
• Docker, Conda, Git, continuous integration systems, or high-performance computing environments.

How it works

You apply on OpenTrain with your resume and then complete the application on the hiring site.

About AI training work

AI training work uses examples, tests, and human reviews to improve how AI systems behave. In this role, your scientific expertise helps evaluate whether AI agents can complete reliable computational workflows, which is why advanced subject knowledge and careful technical review matter.