Senior Agentic AI Engineer — CUDA & HPC

NVIDIA; Santa Clara (CA); On-site; USD 152,000 - 242,000; Full time

Benefits offered by this job

Equity

Comprehensive benefits

Job summary

NVIDIA seeks outstanding Senior High Performance AI Engineers to build the next generation of agentic AI systems for the CUDA ecosystem. You will work across training, architectures, and runtimes, collaborating with NVIDIA teams to deliver scalable, performant solutions.

The role involves designing, building, and optimizing agentic systems, data, environments, and evaluation pipelines, plus co-designing across software and hardware stacks for high-performance workloads.

Qualifications

• Bachelor's degree or equivalent experience in a relevant field.
• 4+ years of AI systems, ML, compilers, HPC experience.
• Hands-on experience with agent systems, coding agents, RL, or modern AI inference systems.
• Strong C/C++, Rust, and Python programming skills with solid software fundamentals.

Responsibilities

• Design, build, and optimize agentic AI systems for the CUDA ecosystem and related architectures.
• Build data, environments, verification, reward, and evaluation systems to improve model and agent capabilities.
• Co-design and optimize agentic systems across NVIDIA's software and hardware stack, from models to GPUs.

Skills

C/C++

Rust

Python

GPU programming

CUDA

Agent systems

Reinforcement learning

Education

Bachelor's degree in Computer Science, Electrical Engineering, or related field

MS or PhD preferred

Tools

TRT-LLM

SGLang

vLLM

Transformer Engine

Job description

NVIDIA seeks outstanding Senior High Performance AI Engineers to build the next generation of agentic AI systems for the CUDA ecosystem. You will work across training, architectures, and runtimes, collaborating with NVIDIA teams to deliver scalable, performant solutions.

The role involves designing, building, and optimizing agentic systems, data, environments, and evaluation pipelines, plus co-designing across software and hardware stacks for high-performance workloads.