You will create and review coding data used to improve large language models. The work focuses on translating CUDA and C++ implementations into Python and checking whether the results preserve the original behavior, numerical results, readability, and performance.
- Translate CUDA and C++ code into equivalent Python implementations with PyTorch and NumPy.
- Analyze CUDA kernels and GPU-accelerated algorithms for structure, efficiency, and function.
- Review generated code translations, debug defects, and write precise corrective feedback.
- Create prompts based on real-world CUDA and PyTorch programming tasks.
- Rank alternative model responses and clearly explain your technical decisions.
- Document coding decisions and suggest improvements based on recurring model errors.
What it pays and takes
This is a fully remote, three-month contractor assignment. The role is open to candidates in India, Pakistan, Nigeria, Kenya, Egypt, Ghana, Bangladesh, Turkey, Brazil, and Mexico.
- Pay: The listing does not provide a rate.
- Time: At least 20 hours per week and at least 4 hours per day.
- Schedule options: 20, 30, or 40 hours per week, with 4 hours of overlap with PST.
- Experience: At least 5 years of overall professional experience, including 3 or more years with Python and 2 or more years with CUDA and C++.
- Technical skills: Strong Python skills for scientific computing with PyTorch and NumPy; knowledge of CUDA programming, C++ fundamentals, GPU computation, and parallelism.
- Language: Strong written and spoken English for technical communication.
- You must be able to analyze CUDA kernels, reproduce their behavior in Python, debug implementations, and maintain numerical consistency across languages.
- Helpful experience includes evaluating AI-generated code or taking part in large language model tuning workflows. A mathematics background may also help with the analytical work.
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 modern artificial intelligence. People prepare examples, write prompts, review model responses, and test code so AI systems can produce more useful and accurate results. In this role, your programming knowledge helps assess whether generated code works correctly and follows the intended logic.
