Top GitHub Contributors Agentic Review Project

You will create and evaluate benchmark tasks that test how well AI systems reason through difficult, real-world problems. The work combines Python development, software testing, dataset preparation, and analysis of model behavior.

You will also study agent trajectories, meaning the steps an AI system takes while working, to find repeated reasoning failures and patterns in where performance drops.

• Design benchmark problems focused on specific reasoning and problem-solving failures.
• Write clear task instructions and produce a correct solution for each task.
• Prepare datasets, Python libraries, metadata, environments, and test notebooks.
• Create rigorous tests and check whether model outputs meet task requirements.
• Evaluate model performance across benchmark tasks.
• Classify logical reasoning failures and identify recurring patterns in agent behavior.

What it pays and takes

This is remote contract work focused on hands-on AI training and model evaluation. Pay details are not provided in the listing.

• Location: Worldwide and fully remote.
• Availability: At least 20 hours per week.
• Language: Fluent English.
• Strong software development experience and GitHub contribution skills.
• Strong expertise in data science, machine learning, coding, or a quantitative technical field.
• Ability to build Python task environments and supporting materials.
• Ability to design reliable tests and assess model outputs against task requirements.
• Helpful background includes benchmark design, model reasoning evaluation, agent behavior analysis, unit tests, reproducible technical environments, or Rust.

How it works

Apply on OpenTrain with your resume, then complete the application on the hiring site.

About AI training work

AI training is the human work behind systems that learn from examples, including writing tests, reviewing model responses, and evaluating how AI solves problems. OpenTrain helps people find these projects, build a profile, and apply in one place.