Materials Engineer & Python Expert - Freelance AI Trainer

About the opportunity

This project-based opportunity connects domain specialists with AI evaluation work for major technology companies. The focus is on assessing, stress-testing, and refining AI systems rather than holding a permanent role.

As a contributor, you will create computational materials science challenges that are tough for a frontier AI model to solve. Each task must have a verifiable answer that can be checked by code, and it must rely on a specialized tool such as ObsPy, instaseis, pyrocko, MITgcm, flopy/MODFLOW, GeoPandas, xmitgcm, or similar software. Simple data cleaning or toy examples will not be sufficient.

Every assignment runs in a locked Linux environment with the relevant tool already installed, and the answer is evaluated by an automated judge.

What you will do

You will act as an expert task author by building problems around the capabilities of a chosen tool. These tasks may involve waveform processing, geophysical inversion, subsurface flow simulation, or validated scientific data workflows.

You will also write a Python reference solution, provide any necessary input files and domain definitions, choose the correct numerical result, and define the tolerance needed for the solution to be accepted.

After creating the task, you will test it across multiple parallel attempts, adjust the difficulty, and continue refining it until the model succeeds only occasionally. Once the task meets the target quality and difficulty range, a senior reviewer in the relevant domain will review it and share feedback.

Calibration is an iterative process that takes patience. The goal is to tune tasks so the pass rate lands in the 10% to 30% range. That may require rewriting waveform setups, tightening inversion settings, and adjusting solver tolerances while observing how the model behaves. Over time, this builds both deeper expertise in the selected tool and practical insight into how frontier models handle seismic, oceanographic, and subsurface flow problems.

How the process works

The workflow is: apply, complete qualification steps, join a project, finish tasks, and receive payment.

Work expectations

During active project periods, the expected commitment is about 10 to 20 hours per week. This is only an estimate and not a guaranteed workload, and it applies solely while the project is running.

Compensation

Contributors may earn up to the equivalent of USD 35 per hour on this project, depending on expertise and pace. Pay can differ from one project to another based on scope, complexity, and the level of specialized knowledge required.

Application note

Submit your CV in English and include your English proficiency level.

Eligibility and fit

This role is suitable for material scientists and engineers with Python experience who are open to part-time, non-permanent project work. Candidates should ideally have a degree in Material Science or a related discipline, at least 2 years of research, applied, or teaching experience, and strong written English at C1 level or above.

You should be comfortable writing Python reference solutions and willing to learn, or already know, at least one scriptable package such as ObsPy, instaseis, pyrocko, MITgcm, xmitgcm, flopy/MODFLOW, or GeoPandas. The role also requires the ability to design problems that genuinely depend on a specialist solver.

Applicants without prior experience with the listed tools may still apply if they are prepared to learn independently and begin contributing quickly.