NVIDIA
Senior Math Libraries Engineer - LLM Integration and Developer Experience
$152K–$288K
US, CA, Santa Clara +5 · Senior · Full-time · Posted Sep 15, 2026
- Location
- US, CA, Santa Clara
US, TX, Remote
US, NY, Remote
US, WA, Remote
US, CA, Remote
US, MA, Remote - Experience
- 3+ years of experience.
- Education
- PhD or MSc in Computational Science, Computer Science, Applied Math, or a related science or engineering field preferred (or equivalent experience).
About the job
Leads the design and delivery of LLM-friendly CUDA-X math library APIs, documentation, tools, and metrics so developers and AI agents can build high-performance applications.
What you'll do
- Architect and implement introspectable, explainable APIs and documentation structures designed for LLMs and AI agents.
- Coordinate with internal and external stakeholders on LLM-enhanced library releases; build metrics, tools, and processes to assess code-generation impact and quality.
- Prototype tools to transform math library APIs into LLM-friendly representations for popular code-generation tools.
What they ask for
- Robust knowledge of LLMs, fine-tuning, RL, RAG, MCP, and agent tooling.
- Experience designing clear, composable APIs and writing high-quality, well-documented code for complex technical domains.
- Ability to prioritize multiple projects and work independently with minimal direction.
- Excellent collaboration, communication, and documentation habits.
Nice to have
- Prior work in AI-assisted software engineering, code generation, or programming language design.
- Familiarity with CUDA-X math library APIs such as cuBLAS, cuFFT, cuSOLVER, and cuSPARSE.
Pay
152,000 USD - 241,500 USD base salary (Level 3); 184,000 USD - 287,500 USD base salary (Level 4)
Tools
CUDA-X, LLMs, RL, RAG, MCP, cuBLAS, cuFFT, cuSOLVER
AI focus
Integrates CUDA-X math libraries into LLM and agentic AI workflows, including LLM-first APIs and documentation, code-generation tooling, and measurement of generated-code impact and quality.
Model-written summary · the original posting is authoritative