Physical Sciences Expert - Terminal Bench
Core
Develop realistic, terminal-based scientific computing tasks for AI benchmarking and evaluation by translating authentic workflows from physics, chemistry, materials science, astronomy, and computational science into reproducible environments.
Role type
Senior IC physical sciences expert (computational modeling & benchmark design)
Builds
Reproducible terminal-based benchmark environments, scientific inputs, executable reference solutions, and automated grading criteria for AI agents
Domain
Physical sciences (physics, chemistry, materials science, astronomy) + computational science
Deliverable
production ML models
Required skills
Advanced academic/research expertise in physical sciences, Python/C/C++/Julia/Bash programming, numerical methods, scientific simulations, computational modeling, Linux/terminal development, reproducibility engineering
Preferred skills
Scientific computing libraries (NumPy, SciPy, pandas, Matplotlib, SymPy, JAX), molecular dynamics, quantum chemistry, finite-difference methods, Monte Carlo simulations, HPC systems, Docker, Conda, Git, CI/CD, automated testing frameworks
Technologies
Python, C/C++, Julia, Bash, NumPy, SciPy, pandas, Matplotlib, SymPy, JAX, OpenMM, ASE, RDKit, Psi4, LAMMPS, GROMACS, Docker, Conda, Git
Responsibilities
Design multi-step terminal-based tasks based on authentic research workflows, build self-contained reproducible computational environments, create datasets and simulation parameters, implement reference solutions, develop automated tests for numerical accuracy and physical consistency, design tasks involving simulations and data fitting, define units and boundary conditions, ensure reliable execution without undocumented dependencies, troubleshoot numerical precision and solver instability, document scientific assumptions and methods
Seniority
Senior, hands-on IC
