The work
You will help create and improve coding datasets used to train and benchmark large language models. The work combines practical software engineering with structured reviews of how well AI models handle real development tasks.
You will write solutions, assess model outputs, build verification tools, and explain your judgments clearly in writing and conversation.
- Review AI-generated code for correctness, efficiency, scalability, reliability, and maintainability.
- Create code examples and precise solutions for AI model training projects.
- Work with Python, JavaScript including ReactJS, C/C++, Java, Rust, and Go.
- Build agents and automated checks that verify code quality and find recurring error patterns.
- Evaluate model performance across prototyping, architecture, API design, production implementation, launch, experimentation, monitoring, and operational maintenance.
- Work with cross-functional teams to improve AI coding solutions against industry benchmarks.
What it pays and takes
The role is a contractor engagement with flexible participation. Pay details are not provided in the role information.
- 20+ hours per week, with partial PST overlap required.
- Worldwide opportunity; professional fluency in English is required.
- The initial engagement lasts one month, with potential extensions based on performance and fit.
- Several years of professional software engineering experience, including at least two continuous years of full-time experience at a top-tier product company.
- Strong full-stack application development and production deployment experience.
- Deep knowledge of software architecture, design, development, debugging, and code quality assessment.
- Ability to provide clear, structured written and oral evaluation rationales.
- Proficiency in one or more of Python, JavaScript, ReactJS, C/C++, Java, Rust, and Go.
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 generate and understand code, text, images, and other content. Contributors review examples, write solutions, and rate model outputs so AI systems learn to produce more useful and reliable results.