About the job
NVIDIA's AI Developer Tools organization is seeking a Senior Research Engineer to join our Research team, where we build the AI coding agents, models, datasets, and evaluations at the heart of NVIDIA's strategy to put AI-powered coding tools in the hands of every CUDA developer. The AI coding space is moving faster than any of us have seen — products rise and fall in months — and we work on it from the epicenter: the company whose hardware most of the AI industry runs on. Our team is small, in-person, and high-velocity. We prototype and ship novel coding agents, fine-tune and evaluate code LLMs, publish benchmarks like ComputeEval, and contribute datasets that feed NVIDIA's Nemotron foundation models. We make NVIDIA's core developer tools — including Nsight Compute and Nsight Systems — first-class citizens for AI agents through MCP servers and Agent Skills. The space shifts every few weeks, and we move with it. In this role, you'll bring applied AI research depth to a team that values shipping as much as experimentation. You won't be filling a narrow gap — you'll pick up significant projects across our portfolio, help set direction on new ones, and partner closely with product teams turning our research into features used by NVIDIA developers and external customers. For experienced AI-for-code practitioners who want to do frontier applied work with the stability and resources of NVIDIA behind them, this is a rare seat.
Responsibilities
Build and improve novel coding agents that help NVIDIA developers write, optimize, and maintain CUDA code — and that work alongside other AI agents in the developer's workflow
Design and ship evaluations, including extensions of our public ComputeEval benchmark, that measure what really matters in AI-powered CUDA development
Fine-tune and specialize code LLMs, and partner with the Nemotron team on the datasets and evaluations that feed NVIDIA's foundation models
Develop Agent Skills, MCP servers, and other tool-use interfaces that make NVIDIA's developer tools (Nsight Compute, Nsight Systems, and more) first-class for AI agents
Generate, curate, and validate synthetic training and evaluation data for CUDA programming
Deliver "net new knowledge" to frontier LLMs through RAG and skill-based systems that keep models current with NVIDIA's fast-moving software stack
Collaborate with partner product teams to turn research prototypes into shipping features used inside NVIDIA and by external customers
Qualifications
Minimum
B.S. in Computer Science or related technical field or equivalent experience (M.S. or Ph.D. a plus)
12+ years of industry experience in applied AI/ML, with meaningful recent work in the AI-for-code space — coding agents, code LLMs, AI developer tools, or adjacent systems
Strong proficiency in Python and sound software engineering practices
Hands-on experience fine-tuning or evaluating LLMs, with appropriate humility about the complexity of training and data work
Fluency with the systems side of LLM-powered agents, including practical concerns like context management, prompt caching, tool-use design, MCP, and Agent Skills
Experience designing or contributing to rigorous evaluations for code generation or agentic systems
Track record of shipping — taking work past the prototype stage and into the hands of real users
Comfortable in a small, collaborative, in-person team with fast direction changes and little process overhead
Preferred
Public contributions in the AI-for-code space — open-source agents or tools, widely-used benchmarks, influential papers, or blog posts with demonstrated real-world impact
Experience building coding agents or code LLMs that real users rely on every day
Familiarity with CUDA or other GPU programming, and/or with NVIDIA profiling tools (Nsight Compute, Nsight Systems) and libraries (cuDNN, cuBLAS, Thrust, CUB)
Experience with synthetic data generation and quality validation for code
Track record of zero-to-one product work, or work inside a recently-rebooted org with a strong mandate and customer pull