Machine Learning Infrastructure Engineer Intern

Responsibilities:

• Identify Training Bottlenecks: Profile and analyze Bird's Eye View (BEV) model training pipelines to pinpoint computational and memory bottlenecks.
• Develop Custom Kernels: Design and implement high-performance custom compute kernels using CUDA, Triton, or C++ to accelerate the model training process.
• Leverage LLMs for Optimization: Explore and integrate Large Language Models (LLMs) to assist in generating high-performance code and optimizing kernel logic.
• Automate Profiling Workflows: Build systems to automate performance profiling and analysis using tools like NVIDIA Nsight and the PyTorch Profiler.
• Iterative Performance Tuning: Continuously analyze profiling data generated by both human and LLM-assisted workflows to maximize GPU utilization and reduce training times.

Required Skills:

• Systems Programming: Strong proficiency in C++ and a solid understanding of memory management, computer architecture, and parallel processing principles.
• Deep Learning Frameworks: Hands-on experience with PyTorch, specifically understanding custom operations, autograd, and training loops.
• Performance-Oriented Mindset: Strong problem-solving skills with a deep interest in performance tuning, algorithmic efficiency, and low-level system optimization.

Preferred Skills:

• GPU Programming Experience: Practical experience writing and optimizing custom GPU kernels using CUDA or OpenAI Triton.
• Hardware Profiling Tools: Familiarity with hardware and software profiling tools, particularly NVIDIA Nsight (Systems/Compute) and the PyTorch Profiler.
• LLM for Code Generation: Experience using or prompting LLMs for code writing, refactoring, or exploring AI-assisted software development workflows.
• Autonomous Vehicle Perception: A foundational understanding of Bird's Eye View (BEV) models, 3D perception, or spatial transformers used in autonomous driving.