GPU Programming Software Engineer

Design, implement, and optimize GPU-based software and kernels used to train large language models, with a strong focus on runtime performance and GPU architecture. You will profile and tune kernels and shaders, build host-side integrations in C++, and deliver GPU tasks that meet project specifications for LLM training and evaluation.

micro1 is an AI data lab that produces training data and evaluations for frontier models and AI agents. Experts from diverse domains, including finance, healthcare, and STEM engineering, contribute real-world knowledge to improve how AI systems learn and reason. micro1 identifies and vets top talent through an AI-driven recruiter and aims to enable 1 billion people to do meaningful work by applying their expertise to AI.

• Design, implement, and optimize GPU-based software solutions using CUDA, WebGPU, or GLSL.
• Profile and fine-tune GPU kernels and shaders for maximum performance and efficiency on NVIDIA-class hardware.
• Develop host-side logic and integrations in C++ to support GPU workloads and training pipelines.
• Translate GPU-focused algorithms into repeatable, spec-compliant task submissions used for LLM training.

• Required skills: CUDA, WebGPU, GLSL, and C++.
• Preferred experience with GPU programming languages and toolchains that run on NVIDIA chips.
• Background in graphics programming, machine learning acceleration, scientific computing, high performance computing, or other GPU-intensive domains.
• Strong proficiency writing and debugging C++ host code that integrates with GPU kernels.

• Role type: Contractor, remote work.
• Compensation is output-based, with experts paid per task that meets project specifications. Time to complete tasks varies by experience and workflow.
• Minimum submission requirements apply, the listing states: ''Experts must submit a minimum of tasks per week''.

Rate range: $60.00 to $85.00 per hour. Compensation is structured as pay-per-task when tasks meet the stated project specifications, and total earnings will depend on the number and complexity of accepted tasks.

Roles are typically filled within 48 hours. Candidates should be ready to begin immediately, as selected experts are expected to start their first tasks within 24 to 48 hours after completing onboarding.