Short contract turning physics, chemistry or materials workflows into reproducible computational tasks that AI agents must solve in a Linux terminal. Suits PhD-level scientists who can build, debug and validate numerical workflows.
Help evaluate how AI systems solve real scientific problems. In this short-term project, you will turn physics, chemistry, materials science or related research workflows into challenging, reproducible tasks that an AI agent must complete in a Linux terminal.
What you'll do
- Design multi-step tasks involving scientific modeling, simulations, quantitative analysis or computational experiments.
- Build working solutions and self-contained environments, prepare scientific inputs, and create automated tests for accurate and reproducible results.
- Debug calculations and dependencies, and document units, assumptions, numerical tolerances and scientific limitations.
- Debug calculations and dependencies, and document units, assumptions, numerical tolerances and scientific limitations.
- A PhD, postdoctoral experience or equivalent advanced technical experience in a relevant physical-science discipline.
- Strong programming skills in Python, C/C++, Julia, Bash or another scientific programming language, plus experience working in Linux or terminal environments.
- Experience independently creating and validating scientific workflows, with an understanding of numerical precision, physical constraints and reproducibility.
- Benchmark design, research software, publications and open-source contributions are welcome but not required.
- Benchmark design, research software, publications and open-source contributions are welcome but not required.
- Expected project length: five weeks.
- Availability: 40 hours per week, including four hours of overlap with Pacific working hours each day.
- Selection includes a short interest form, a 60-minute assessment and a final review with the team.
- Selection includes a short interest form, a 60-minute assessment and a final review with the team.
- $300 per approved task.
- Tasks are estimated to take six to eight hours on average.
Listing from Braintrust's job board. Details can change; the platform's page is the final word.
