Zyphra
Research Engineer - AI Performance & Kernel Optimization
San Francisco · On-site · Full-time · Posted Mar 16, 2026
- Education
- Background in a highly technical field such as physics, mathematics, theoretical computer science, computer science, or electrical engineering.
- Visa sponsorship
- Relocation and immigration support on a case-by-case basis.
About the job
Optimize kernels and large-scale training and inference systems to improve throughput, latency, and accelerator utilization for language models.
What you'll do
- Develop and tune kernels and performance across GPUs and other accelerators.
- Profile and remove bottlenecks in memory movement, communication, scheduling, and compute utilization.
- Optimize distributed training and inference for large MoE models and collaborate with research and infrastructure teams.
What they ask for
- Strong engineering aptitude for building reliable, high-performance systems.
- Strong low-level performance intuition and ability to reason about hardware-software interactions.
- Experience writing performant GPU kernels using PTX, CUDA, HIP, Triton, or other kernel DSLs.
- Experience optimizing ML workloads for large-scale training; understanding of distributed training systems and parallelism schemes.
Nice to have
- Experience with non-NVIDIA accelerators, such as AMD, AWS Trainium, Google TPU, Qualcomm, ARM, Intel, and custom ASICs.
- HPC experience.
Tools
PTX, CUDA, HIP, Triton, AMD, MI300x, MI355x
AI focus
Performance optimization for large-scale language-model training and inference, including distributed systems for large MoE models.
Model-written summary · the original posting is authoritative