Pod lead - Physics (Terminal Bench Science)

Role Overview

We are seeking an experienced Physics expert to lead a pod of 5–10 trainers developing realistic, terminal-based scientific and technical tasks for Terminal Bench Science.

You will review every task your trainers produce for physical correctness, reproducibility, and grader reliability, and coach trainers to consistently deliver high-quality, challenging tasks.

What you'll do:

• Review trainer tasks end to end, including task statements, datasets, simulation inputs, environments, reference solutions, and automated tests.
• Verify that tasks reflect authentic physics workflows and require genuine multi-step reasoning.
• Check that tools, libraries, and dependencies are pinned and that results are reproducible.
• Validate dimensional consistency, unit systems and conventions, conservation laws, symmetries, integrator energy drift, convergence, and statistical treatment of uncertainties.
• Ensure tolerances reflect legitimate stochastic and numerical variation without accepting physically incorrect results.
• Ensure graders verify observables, fitted parameters, uncertainties, or physical invariants rather than formatting.
• Ensure tasks cannot be gamed through hard-coded outputs or bypassed computation.
• Provide clear, actionable written feedback and track revisions through closure.
• Onboard and mentor trainers.
• Allocate work, track quality and throughput, and escalate blockers.
• Maintain pod documentation covering guidelines, recurring issues, and exemplary tasks.

What we're looking for:

• Ph.D., postdoctoral experience, or equivalent advanced experience in physics.
• Strong programming skills in Python, C/C++, Julia, or Fortran, with Linux proficiency.
• Hands-on experience with computational physics, simulation, or physics data analysis.
• Proven experience reviewing code, analyses, or research outputs.
• Experience mentoring or leading small technical or research teams.
• Excellent written communication skills for providing precise, constructive feedback.

Preferred Qualifications:

• Familiarity with QuTiP, Qiskit, Kwant, ITensor, ROOT, Geant4, Astropy, Dedalus, MEEP, or similar tools.
• Familiarity with Docker, Git, CI systems, and automated testing.
• Experience designing benchmarks or simulation-based evaluations.

Bonus Points:

• Experience evaluating AI coding or terminal agents.
• Experience working in large collaborations with formal internal review processes.
• Prior experience in AI training or evaluation programs.

Offer Details:

• Remote
• Full-time dedication (40 hours/week) - 8 hours per day with an overlap of 4 hours with PST.
• Location: Bangladesh, Brazil, Colombia, Egypt, Ghana, India, Indonesia, Turkey
• Engagement Length: 3 weeks

Important note: Due to the high volume of applications, we are unable to provide individual feedback at this stage. Only candidates shortlisted for the next steps will be contacted by our team.