MAX Kernel Engineering Manager

Modular•United States,

11d•$248,000 - $372,000•Hybrid

About The Position

The MAX Kernel team is the performance foundation of Modular's inference tech stack (MAX), which is built to run GenAI models fast across various hardware like NVIDIA, AMD, Qualcomm, Trainium, and TPU. The team develops kernels using Mojo, GEMM, attention (MHA/MLA/MSA), MoE routing, and grouped matmul, along with portable tile abstractions that enable a single kernel to lower to multiple targets. As the Engineering Manager of the MAX Kernel team, you will lead a team of kernel engineers, own the kernel roadmap, and be responsible for per-model x per-hardware performance optimization against the best vendor and open-source stacks. You will collaborate closely with the Framework, Serve, Compiler/Runtime, and Hardware Enablement teams, and contribute to growing a kernel engineering community that spans Modular and Qualcomm.

Requirements

3+ years as an engineering manager leading GPU kernel, compiler, HPC, or ML performance engineering teams.
Strong cross-functional communication skills, capable of driving technical decisions with tech leads and communicating status, risks, and tradeoffs to engineering leadership and customers.
Proven track record of hiring, retaining, and growing senior kernel or performance engineers.
Working knowledge of writing and optimizing GPU/accelerator kernels (CUDA, HIP/ROCm, Triton, CUTLASS/CuTe or similar).
Deep understanding of profiling, benchmarking, and roofline analysis.

Nice To Haves

Experience with Mojo or other kernel and tile-level programming models (Triton, TileLang, CuTe).
Multi-target experience beyond NVIDIA, including AMD, NPUs/ASICs, or edge/on-device.
Experience with AI-assisted or agentic kernel development.
Open-source contributions to kernel libraries or inference engines.

Responsibilities

Lead, hire, coach, and grow a team of approximately 12 kernel engineers across NVIDIA, AMD, and ASIC hardware, managing career development, performance, and team health.
Own the MAX kernel roadmap and delivery, ensuring performance matches or exceeds open-source kernel libraries on essential business models and hardware targets.
Support portable kernel architecture (TileTensor / TensorEngine / TileIO, reusable building blocks like MegaFFN and the attention family) in collaboration with tech leads.
Partner cross-functionally with Framework, Serve, Compiler/Runtime, Hardware Enablement, and Qualcomm kernel teams on priorities, interfaces, and escalations.
Advance AI-assisted kernel development (kernel agents, fuzz verification, playbooks) and foster the kernel engineering community across organizational boundaries.