Kernel Engineer

About the Company

We're building autonomous research agents for recursive self-improvement (multi-agent systems that propose, run, and analyze machine learning experiments). We're a small team based in San Francisco.

Responsibilities

• Write and optimize GPU kernels (CUDA, ROCm, Triton, or similar) for training and inference workloads including attention variants, MoE layers, and custom activations
• Profile real workloads with hardware counters and translate findings into specific kernel-level optimizations
• Co-design kernels with research teams to align kernels and algorithms
• Integrate optimized kernels into training and serving stacks and verify end-to-end performance wins
• Maintain kernel quality as hardware, frameworks, and workloads shift
• Share kernel-level expertise across the team

Requirements

• 4+ years writing performant GPU kernels (CUDA, ROCm, Triton, or production-grade equivalent)
• Hardware-level fluency in memory hierarchy, occupancy, register pressure, tensor cores, and warp scheduling
• Proficiency with profiling tools like Nsight or ncu
• Proven track record of shipping kernel-level optimizations that impact measurable metrics
• Strong systems expertise regarding how kernels integrate into larger frameworks
• Ability to read framework-level Python and C++

Preferred Qualifications

• Open-source contributions to kernel libraries, compilers, or ML frameworks
• Experience with multiple accelerator architectures, preferably AMD GPUs
• Familiarity with collective communication primitives like NCCL
• Background in compilers or runtimes