Develop and optimize numerical algorithms and Device eXtension (Dx) APIs for GPU-accelerated math libraries. Work with the libraries team on NVIDIA GPUs, performance analysis, and tooling for LLM-friendly API representations.
Pursuing an MS or PhD degree in Computer Science, Artificial Intelligence, Applied Mathematics, or a related field.
Strong C++ programming, debugging, performance analysis, and test-building skills.
Parallel or GPU programming experience using CUDA, OpenMP, OpenCL, SYCL, or similar.
Experience with linear algebra, FFT, or data compression algorithms.
Collaboration, communication, and documentation habits.
Nice to have:
Understanding of composability and fusions, compilers, and programming-language implementation.
Knowledge of GPU and network hardware architecture.
Familiarity with Python and recent CUDA developments such as cuTile and CUDA Tile C++.
Prior work in AI-assisted software development; knowledge of LLMs, MCP, and building RAGs or agenting tooling.
Exposure to numerical methods, floating-point arithmetic, and numerical error analysis.
Prototype and develop numerical algorithms for high-performance math device extension libraries: cuBLASDx, cuSolverDx, cuFFTDx, cuRANDDx, nvCOMPDx, NPPDx, or libmathdx on NVIDIA GPUs.
Analyze the performance of GPU implementations and identify opportunities for improvement.
Collaborate with team members to understand software use cases and requirements.
Prototype tooling and solutions to transform math libraries’ APIs into LLM-friendly representations across popular code-generation tools.
