AI Frameworks Engineer

Intel's Data Center Group (DCG) develops AI frameworks and workloads for large language model (LLM) inference on Intel GPUs, focusing on high-performance and scalable AI solutions.

• Bachelor's degree in AI, Computer Science, Computer Engineering, Mathematics, Physics, or related field.
• 2-3 years relevant experience with Bachelor's degree, or 1+ year with Master's degree.
• Advanced expertise in Python programming.
• Proficiency in PyTorch, including its internals or extension mechanisms.
• Experience with AI inference and large language models (LLMs).
• Strong problem-solving skills with debugging and designing optimized software and algorithms.
• Proficiency in Linux user environments.
• Solid foundation in AI frameworks, machine learning algorithms, and software engineering.

Preferred Qualifications:

• Experience writing or optimizing GPU compute kernels (e.g., Triton, CUDA, SYCL, OpenCL).
• Experience in performance profiling and hardware-level optimization of AI workloads.
• Familiarity with parallel and distributed computing and collective communication libraries.
• Familiarity with vLLM or other LLM inference serving frameworks.
• Track record of contributions to major open-source AI projects.
• Understanding of numerical precision and quantization formats (FP8, INT8, INT4) and accuracy implications.
• Creativity in solving technical and system-level challenges.
• Strong communication skills for global team collaboration.

• Design, develop, and optimize AI framework software for LLM inference on Intel GPUs, contributing upstream to open-source frameworks such as vLLM and PyTorch.
• Develop and optimize compute kernels (e.g., Triton, SYCL) for attention, mixture-of-experts, quantization, normalization, and other transformer inference components.
• Implement and tune distributed execution paths including tensor, pipeline, and data parallelism, collective communication, and asynchronous data movement.
• Transform and optimize computational-graph representations and develop machine learning primitives in mathematical and kernel libraries.
• Profile inference workloads to identify bottlenecks and collaborate with component teams on solutions.
• Optimize code for diverse hardware backends adapting to memory hierarchy, instruction set, and execution models.
• Build and extend unit tests, correctness checks, and performance regression coverage.
• Collaborate with framework, compiler, and machine learning engineers globally and engage with the open-source community through code review and technical discussions.

All qualified applicants will receive consideration without regard to protected characteristics by local law. Job posting details (work model, location, time type) are subject to change.