Mathematics & Python Expert - Freelance AI Trainer at Mindrift

, , Poland -

Full Time

Start Date

Immediate

Expiry Date

18 Sep, 26

Salary

45.0

Posted On

21 Jun, 26

Experience

2 year(s) or above

Remote Job

Yes

Telecommute

Yes

Sponsor Visa

Skills

Mathematics, Python, Z3, Cvc5, SageMath, Macaulay2, Singular, CasADi, IPOPT, SDPB, G6K, Fpylll, AI Training, Problem Design, Linux, English Proficiency

Industry

technology;Information and Internet

Description

Please submit your CV in English and indicate your level of English proficiency. 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. What this opportunity involves You design mathematics 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 Z3, cvc5, SageMath, Macaulay2, or others. NumPy or SymPy on their own won't cut it. Each problem runs inside a sealed Linux container with the tool pre-installed and a programmatic judge that grades the model's answer. As an expert author, you: • Pick an anchor tool and design a problem that hinges on its usage. • Write a Python reference solution, supply input files optionally where needed. • Decide the numerical answer and how close the model needs to get 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. They will provide 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, re-tightening, and watching how the agents act. You'll learn how these agents cut corners, where it stalls, where it converges. This time compounds in two directions. You come out of each task with deeper command of the anchor tool itself, and also get a hands on working intuition for how a frontier model navigates complex scientific problems. What we look for This opportunity is a good fit for mathematicians with an experience in python open to part-time, non-permanent projects. Ideally, contributors will have: • Degree in Mathematics (Pure or Applied) 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 mathematical package: Z3, cvc5, Macaulay2, Singular, CasADi, IPOPT, SDPB, G6K, fpylll, or SageMath; • Ability to design problems that genuinely require a specialized solver; • 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. How it works Apply → Pass qualification(s) → Join a project → Complete tasks → Get paid Project time expectations For this project, tasks are estimated to require around 10–20 hours per week during active phases, based on project requirements. This is an estimate, not a guaranteed workload, and applies only while the project is active. Compensation On this project, contributors can earn up to $45 per hour equivalent, depending on their level and pace of contribution. Compensation varies across projects depending on scope, complexity, and required expertise. Please note that other projects on the platform may offer different earning levels based on their requirements.

How To Apply:

Incase you would like to apply to this job directly from the source, please click here

Responsibilities

Design complex mathematics problems that require specialized solvers to challenge and improve frontier AI models. Develop Python reference solutions and iteratively tune problem difficulty based on model performance.