Inference & GPU Kernel Engineer

Touchdown Labs, Inc.; Santa Clara (CA); On-site; USD 200,000 - 270,000; Full time

Job summary

Touchdown Labs is hiring an Inference & GPU Kernel Engineer in Santa Clara, CA to accelerate customer models with faster, cheaper outputs. You’ll profile workloads, identify bottlenecks, and optimize serving configurations, including GPU kernels when gains justify them.

You should bring deep GPU programming experience, strong proficiency in C++ and Python, and hands-on work with CUDA, Triton, or HIP. Knowledge of NVIDIA/AMD accelerators and benchmarking discipline is highly valued.

Qualifications

• Profiling and bottleneck analysis of AI workloads.
• Serving configuration and optimization for inference deployments.
• Experience writing or tuning GPU kernels when beneficial.
• Evidence of improved inference or training performance with profilers.

Responsibilities

• Profile customer workloads and identify bottlenecks.
• Implement serving configuration optimizations.
• Develop GPU kernels when gains justify them.

Skills

GPU programming

Profiling

C++

Python

Tools

CUDA

Triton

HIP

Job description

Careers · Forward deployed & AI platform# Inference & GPU Kernel EngineerMake customer models faster and cheaper per useful output.**Location:** Santa Clara, CAAll roles →Apply for this role →---SeniorSanta ClaraOpenings: 1–2Open role## Inference & GPU Kernel Engineer### Our missionAI is becoming fundamental to how the world works. When it is, the ability to compute, and to manufacture the hardware that computes, should not be concentrated in a few places. Touchdown's mission is to make compute accessible to the whole world and to reindustrialize advanced semiconductor manufacturing in the United States. We co-design everything for next-generation compute: the AI systems people use, the software that serves them, the architectures and chips underneath, and the packaging, test, and manufacturing that make them real. Co-design brings down the complexity and cost of building compute. That is what makes manufacturing in America competitive again, and what puts compute within reach of people and companies everywhere. We build alongside the world's best manufacturing ecosystems, including Taiwan's, and every role here makes one layer better and carries what it learns to the rest.### Why this role mattersInference is where most AI compute is spent. Making it faster and cheaper per useful output is the clearest proof of what co-design across software and hardware can do.### Technical scopeProfiling and bottleneck analysis; serving configuration (batching, KV cache management, speculative decoding, quantization); GPU kernels in CUDA, Triton, or HIP; memory and interconnect behavior; benchmark design with accuracy guards; NVIDIA and AMD accelerators.### First 90 days* Profile customer workloads and name bottlenecks with traces.* Ship measured improvements with reproducible benchmarks.* Hand off so customer teams keep the gains.### What you will own* Profiling and bottleneck analysis.* Serving configuration and optimization.* GPU kernels when the gain justifies them.### What you bring* GPU programming or inference-systems depth.* Profilers and benchmark discipline.* C++ and Python.* Has improved inference or training performance with profiler evidence.* Has written or tuned CUDA, Triton, or HIP kernels.### Helpful experience* NVIDIA and AMD accelerators### How we work* AI-native by default. You use coding agents, AI-assisted analysis, and automation every day, and you check their output against measurements before anyone relies on it.* We work alongside foundries, test houses, OSATs, and research labs, and we build with partners rather than against them.### How the role works* Full-time.* Location: Santa Clara, CA.* Openings: 1–2.* Pay: $200,000–$270,000 base (Santa Clara, CA).* Ranges are base pay for U.S. roles and depend on location, experience, and level. Full-time roles also include equity and benefits. Taiwan pay is set to the local market and discussed in the first call.* Start date and employment details are discussed during the process.Apply for this role →Applications are reviewed against the work described here. We do not use a degree, title, or keyword list as a substitute for evidence.