Physics & Python Expert - Freelance AI Trainer

Mindrift connects specialists with project-based AI opportunities for leading tech companies, focused on testing, evaluating, and improving AI systems. Participation is project-based, not permanent employment.

• Degree in Physics (Theoretical, Experimental, or Computational) or related field
• 2+ years of research, applied, or teaching experience
• Python proficiency for writing reference solutions
• Fluency with — or strong willingness to independently learn — at least one scriptable physics package: FEniCS / DOLFINx, OpenFOAM, Meep, MPB, openEMS, Geant4, PYTHIA8, ROOT / PyROOT, WarpX, REBOUND, MESA, CAMB, CLASS, or bilby
• Ability to design problems that genuinely require a specialized simulation tool
• Strong written English (C1+)

No prior experience with the listed tools? You're still welcome to apply — as long as you're ready to get up to speed on your own and hit the ground running.

You design computational physics problems to challenge a frontier AI model. The problem must have an answer verifiable by code, and the problem has to require a specialized tool like FEniCS, OpenFOAM, Meep, REBOUND, CAMB, or others.

As an expert author, you:

• Pick an anchor tool and design a problem that hinges on its physics models, integrators, Monte Carlo kernels, or PDE discretisations.
• Write a Python reference solution, supply input files and domain or initial condition definitions where needed.
• Decide the numerical answer and how close the model needs to get — with a domain-appropriate tolerance — to count as right.
• Test the problem against the model in batches of parallel attempts, tuning the problem difficulty until the agent only succeeds in a small number of attempts.
• Once you're happy with the task, and it scores within range, the task goes to a senior reviewer in your subfield who provides feedback to ensure task quality is high.

Calibration requires patience. You're tuning the problem against batches of parallel runs of the agent, aiming for a pass rate in the 10–30% band. Reaching that means rewriting field configurations, tightening initial conditions and solver parameters, and watching how the agents act.

Please submit your CV in English and indicate your level of English proficiency.