Description

Own production readiness, performance optimization, and stack validation for AI applications on DGX Station.

This role is on-site.

Responsibilities

• Define readiness criteria and validate AI workloads (NemoClaw, Hermes, NIM) across single- and multi-GPU configurations.
• Profile and optimize LLM/deep learning performance (PyTorch, TensorFlow, JAX) on GB300 Blackwell architectures.
• Identify and resolve bottlenecks in GPU compute, NVLink, PCIe, and CPU-GPU communication.
• Collaborate with framework and compiler teams to improve kernel fusion, graph execution, and memory management.
• Build automated benchmarking infrastructure to track regressions across key models and driver updates.

Required Skills

• 12+ years in systems software engineering with deep learning infrastructure or GPU performance analysis.
• Strong proficiency in C/C++, CUDA, and Python, with ability to modify GPU kernels.
• Deep understanding of PyTorch, TensorFlow, or JAX internals: graph execution, operator dispatch, memory management.
• Experience profiling GPU workloads using Nsight Systems, Nsight Compute, or CUPTI.
• Strong grasp of GPU architecture: compute units, memory hierarchy, NVLink, and multi-GPU scaling.
• Experience with inference optimization including quantization (INT8/FP8), TensorRT, and torch.compile.
• BS or MS in Computer Science, Electrical Engineering, or equivalent experience.

Preferred Skills

• Experience optimizing LLM training/inference on multi-GPU NVIDIA systems (DGX, HGX).
• Contributions to open-source AI frameworks, CUDA libraries, or inference engines.

Key Skills

Education

Any Graduate