GPU-Optimized LLM Inference Engineer

Intel Corporation; Austin (TX); Hybrid; USD 171,000 - 315,000; Full time

Benefits offered by this job

Stock bonuses

Health benefits

Hybrid work model

Job summary

Intel Corporation in the United States (US) seeks an experienced AI Infrastructure Engineer to push LLM inference on Intel GPUs to new performance heights. You will work end-to-end across profiling, kernel development, and upstreaming optimizations into vLLM, SGLang, and PyTorch, influencing hardware and software co-design.

Responsibilities include diagnosing bottlenecks, developing high-performance kernels for attention and MoE, and collaborating with architecture and compiler teams to plan

Qualifications

• 4+ years of software engineering experience in GPU computing, AI systems, or HPC.
• Proficiency in modern C++ and Python.
• Familiarity with AI inference workloads and performance optimization.

Responsibilities

• Own the end-to-end optimization pipeline for running state-of-the-art LLMs on Intel GPUs.
• Profile, diagnose, and resolve cross-stack performance bottlenecks.
• Design, write, and optimize custom high-performance kernels for attention, MoE, quantization, and operator fusion.
• Upstream architectural improvements into vLLM, SGLang, and PyTorch; bridge hardware teams with open-source communities.
• Collaborate with architecture and compiler teams to shape future GPU roadmaps based on GenAI workloads.

Skills

C++

Python

GPU computing

HPC

AI systems

Education

Bachelor's degree in CS/SE/AI/ML

Master's degree

PhD

Tools

Triton

SYCL

CUDA/CUTLASS

PyTorch

vLLM

SGLang

Job description

Intel Corporation in the United States (US) seeks an experienced AI Infrastructure Engineer to push LLM inference on Intel GPUs to new performance heights. You will work end-to-end across profiling, kernel development, and upstreaming optimizations into vLLM, SGLang, and PyTorch, influencing hardware and software co-design.

Responsibilities include diagnosing bottlenecks, developing high-performance kernels for attention and MoE, and collaborating with architecture and compiler teams to plan