KERNEL ENGINEER

What you'll do

• Develop and optimize kernels for workloads such as attention, MoE layers, activations, and communication primitives; co-design kernels and algorithms with researchers.
• Profile real workloads, identify optimizations, and benchmark their end-to-end impact.
• Integrate and maintain optimized kernels in training and serving stacks, and share kernel expertise across the team.

What they ask for

• 4+ years writing performant GPU kernels using CUDA, ROCm, Triton, or a production-grade equivalent.
• Fluency in GPU hardware concepts, including memory hierarchy, occupancy, register pressure, tensor cores, and warp scheduling.
• Experience profiling with Nsight, ncu, or comparable tools, and a track record of measurable kernel-level optimizations in a real system.
• Strong systems expertise and comfort reading framework-level Python and C++.

Nice to have

• Open-source contributions to kernel libraries, compilers, or ML frameworks; compiler or runtime background.
• Experience with multiple accelerator architectures, preferably AMD GPUs; familiarity with NCCL or equivalent.

Tools

CUDA, ROCm, Triton, Nsight, ncu, Python, C++, NCCL

AI focus

The company builds autonomous research agents for recursive self-improvement; this role accelerates their machine-learning training and inference workloads through kernel engineering.