Research Engineer, Infrastructure, Kernels

Thinking Machines Lab Inc.; San Francisco, Northern (CA, KY); Hybrid; USD 350,000 - 475,000; Full time

Benefits offered by this job

Health, dental, and vision benefits

Unlimited PTO

Parental leave

Relocation support

Job summary

Thinking Machines Lab Inc. in San Francisco seeks an infrastructure research engineer to design, optimize, and maintain the compute foundations powering large-scale language model training.

You will develop high-performance ML kernels (CUDA, CuTe, Triton) and improve the distributed compute stack for scalable AI systems. You’ll collaborate with researchers and systems architects, prototype kernel implementations, and profile performance across hardware generations to shape numerical and

Qualifications

• Bachelor’s degree or equivalent experience in a technical field.
• Strong engineering skills with maintainable code and debugging in complex codebases.
• Understanding of deep learning frameworks (e.g., PyTorch, JAX) and their system architectures.
• Ability to collaborate with cross-functional teams and take initiative across stacks.

Responsibilities

• Design and implement custom ML kernels (CUDA, CuTe, Triton) for core LLM operations such as attention and matrix multiplication.
• Design and optimize compute primitives to reduce memory bandwidth bottlenecks and improve kernel efficiency.
• Collaborate with research teams to align kernel optimizations with model architecture goals.
• Develop and maintain a library of reusable kernels and performance benchmarks.
• Contribute to infrastructure stability, reproducibility, and high compute utilization.
• Document insights via talks, papers, or open-source contributions.

Skills

Kernel development

Performance profiling

GPU programming

Collaboration

Strong coding skills

Deep learning systems

Education

Bachelor’s degree in CS/EE/related

Tools

CUDA

CuTe

Triton

PyTorch/JAX

Job description

The mission of Thinking Machines is to build AI that extends human will and judgment.

About the Role

We’re looking for an infrastructure research engineer to design, optimize, and maintain the compute foundations that power large-scale language model training. You will develop high-performance ML kernels (e.g., CUDA, CuTe, Triton), enable efficient low-precision arithmetic, and improve the distributed compute stack that makes training large models possible.

This role is perfect for an engineer who enjoys working close to the metal and across the research boundary. You’ll collaborate with researchers and systems architects to bridge algorithmic design with hardware efficiency. You’ll prototype new kernel implementations, profile performance across hardware generations, and help define the numerical and parallelism strategies that determine how we scale next-generation AI systems.

What You’ll Do

• Design and implement custom ML kernels (e.g., CUDA, CuTe, Triton) for core LLM operations such as attention, matrix multiplication, gating, and normalization, optimized for modern GPU and accelerator architectures.

• Design and think through compute primitives to reduce memory bandwidth bottlenecks and improve kernel compute efficiency.

• Collaborate with research teams to align kernel-level optimizations with model architecture and algorithmic goals.

• Develop and maintain a library of reusable kernels and performance benchmarks that serve as the foundation for internal model training.

• Contribute to infrastructure stability and scalability, ensuring reproducibility, consistency across precision formats, and high utilization of compute resources.

• Document and share insights through internal talks, technical papers, or open-source contributions to strengthen the broader ML systems community.

Skills and Qualifications

Minimum qualifications:

• Bachelor’s degree or equivalent experience in computer science, electrical engineering, statistics, machine learning, physics, robotics, or similar.

• Strong engineering skills, ability to contribute performant, maintainable code and debug in complex codebases

• Understanding of deep learning frameworks (e.g., PyTorch, JAX) and their underlying system architectures.

• Thrive in a highly collaborative environment involving many, different cross-functional partners and subject matter experts.

• A bias for action with a mindset to take initiative to work across different stacks and different teams where you spot the opportunity to make sure something ships.

• Proficiency in CUDA, CuTe, Triton, or other GPU programming frameworks.

• Demonstrated ability to analyze, profile, and optimize compute-intensive workloads.

Preferred qualifications — we encourage you to apply if you meet some but not all of these:

• Experience training or supporting large-scale language models with tens of billions of parameters or more.

• Track record of improving research productivity through infrastructure design or process improvements.

• Experience developing or tuning kernels for deep learning frameworks such as PyTorch, JAX, or custom accelerators.

• Familiarity with tensor parallelism, pipeline parallelism, or distributed data processing frameworks.

• Experience implementing low-precision formats (FP8, INT8, block floating point) or contributing to related compiler stacks (e.g., XLA, TVM).

• Contributions to open-source GPU, ML systems, or compiler optimization projects.

• Prior research or engineering experience in numerical optimization, communication-efficient training, or scalable AI infrastructure.

Logistics

• Location: This role is based in San Francisco, California.

• Compensation: Depending on background, skills and experience, the expected annual salary range for this position is $350,000 - $475,000 USD.

• Visa sponsorship: We sponsor visas. While we can't guarantee success for every candidate or role, if you're the right fit, we're committed to working through the visa process together.

• Benefits: Thinking Machines offers generous health, dental, and vision benefits, unlimited PTO, paid parental leave, and relocation support as needed.