AI Model Evaluator & Challenge Designer (Remote, Part-Time)

OpenTrain AI, Inc.; United States; Remoto; USD 138.000 - 234.000; Tempo parcial

Benefícios oferecidos por esta vaga

Remote work

Resumo da vaga

OpenTrain AI, Inc. is seeking a remote, part-time contractor to create technical problems and grading guides and to assess AI answers on machine learning inference, GPU kernels, inference serving, and large-model deployment.

The work focuses on judging model answers and explaining what strong technical reasoning looks like, rather than building production systems. You will design realistic scenarios, write reference solutions and grading rubrics, review AI-generated responses for accuracy and

Qualificações

• Pay: $100 to $170 per hour.
• Schedule: 20+ hours per week, flexible asynchronous work.
• Location: Fully remote; eligible countries.
• Language: Strong written English for technical explanations, reference answers, grading rubrics, and feedback.

Responsabilidades

• Design realistic technical scenarios and problem sets.
• Write reference solutions and grading rubrics that define correctness, depth, and technical judgment.
• Review AI-generated answers for accuracy, reasoning quality, and understanding of the subject.
• Provide structured feedback and write clear technical explanations of your decisions.

Conhecimentos

Strong written English

Technical judgment

Feedback writing

Formação

PhD or doctorate

Master's degree with strong research record

Ferramentas

SGLang

vLLM

Mamba/Mamba2

TensorRT-LLM

GPU kernel development

Descrição da vaga

OpenTrain AI, Inc. is seeking a remote, part-time contractor to create technical problems and grading guides and to assess AI answers on machine learning inference, GPU kernels, inference serving, and large-model deployment.

The work focuses on judging model answers and explaining what strong technical reasoning looks like, rather than building production systems. You will design realistic scenarios, write reference solutions and grading rubrics, review AI-generated responses for accuracy and