GPU Kernel Engineer

Sciforium•San Francisco, CA

23h•$190,000 - $250,000

About The Position

Sciforium is an AI infrastructure company developing multimodal AI models and a high-efficiency serving platform. The GPU Kernel Engineer will design and optimize custom GPU kernels for large-scale AI systems, working across the hardware and software stack—from low-level kernel development to integrating optimized operations into machine-learning frameworks used for training and inference. The role focuses on GPU programming, systems engineering, and AI workloads, with the goal of improving the efficiency and scalability of Sciforium’s ML platform.

Requirements

At least 5 years of industry or research experience in GPU kernel development or high-performance computing.
Bachelor’s, Master’s, or PhD in Computer Science, Computer Engineering, Electrical Engineering, Applied Mathematics, or a related field.
Strong programming skills in C++ and Python, and familiarity with machine-learning frameworks.
Deep expertise in CUDA/ROCm, GPU memory models, and performance optimization strategies.
Hands-on experience with Triton and/or JAX Pallas for custom kernel development.
Strong understanding of PTX, GPU assembly, and low-level GPU execution.
Extensive experience writing and optimizing custom GPU kernels in C++ and PTX.
Proven ability to integrate low-level kernels into PyTorch, JAX, or similar frameworks.
Experience working with large-scale LLM training or inference workloads.

Nice To Haves

Experience with AMD GPUs and ROCm optimization.
Familiarity with JAX FFI and custom ML operator development.
Experience with efficient model-serving frameworks such as vLLM or TensorRT.
Experience with TPUs, XLA, or similar accelerator programming environments.
Contributions to open-source ML systems, compilers, or GPU kernels.

Responsibilities

Design, implement, and optimize custom GPU kernels using C++, PTX, CUDA, ROCm, Triton, and/or JAX Pallas.
Profile and optimize end-to-end ML operation performance, particularly for large-scale LLM training and inference.
Integrate low-level GPU kernels into PyTorch, JAX, and custom internal runtimes.
Develop performance models, identify bottlenecks, and deliver kernel-level improvements that accelerate AI workloads.
Collaborate with ML researchers, distributed systems engineers, and model-serving teams to optimize compute performance across the stack.
Work with hardware vendors, including NVIDIA and AMD, and stay current with GPU architecture capabilities and compiler and toolchain improvements.
Contribute to tooling, documentation, benchmarking suites, and testing frameworks to support correctness and reproducible performance.