Mathematics & Python Expert - Freelance AI Trainer

Overview

This role is part of a project-based AI program that pairs subject-matter specialists with leading technology companies to help test, assess, and improve AI systems. It is not a permanent position; instead, contributors work on individual projects as needed.

What the work involves

You will create mathematics challenges designed to push a frontier AI model. Each task must have a verifiable answer that can be checked by code, and it should rely on a specialized mathematical or computational tool rather than general-purpose packages alone. Examples of suitable tools include Z3, cvc5, SageMath, Macaulay2, and similar software. Every task is executed in a locked Linux environment with the required tool already installed and a programmatic judge that evaluates the model’s response.

Core duties

• Select an appropriate anchor tool and build a problem that truly depends on it.
• Prepare a Python reference solution and, when needed, provide input files.
• Determine the correct numerical answer and define the acceptable margin of error for scoring.
• Run the task against multiple parallel model attempts, then refine the problem difficulty until only a small share of attempts succeed.
• After calibration, submit the task for senior review in the relevant specialty and incorporate feedback to improve quality.

Calibration and task tuning

Task calibration takes patience because the aim is to reach an approximate 10% to 30% pass rate. This often means rewriting the task, tightening the constraints, and observing how the model responds. Over time, this process improves your understanding of both the anchor tool and the ways frontier models approach complex scientific work.

Who this suits

This opportunity is well suited to mathematicians with Python experience who are open to part-time, non-permanent project work. Applicants should be comfortable working independently and building tasks that require specialist tools to solve properly.

Application requirements

Please send your CV in English and clearly state your English proficiency level.

How the process works

Application steps are: apply, complete one or more qualification stages, join a project, complete tasks, and receive payment.

Working time

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

Compensation

Contributors may earn up to the equivalent of USD 35 per hour, depending on their level and pace of work. Pay can vary across projects based on scope, complexity, and the expertise required.