Math PhD + NVIDIA NIM/CUDA Engineer

Our team is tackling high-complexity challenges in computational modeling and GenAI-driven analytics. We are currently building a next-generation acceleration layer for predictive simulations that requires a blend of deep theoretical knowledge and low-level GPU optimization. You will be responsible for ensuring our mathematical models don’t just work they run at the theoretical limits of available hardware.

Requirements

• PhD in Computer Science, Physics, Mathematics, or a related quantitative field with a focus on high-performance computing or numerical methods
• 3+ years of experience in Python Engineering (Middle/Senior level) with a deep understanding of asynchronous programming and system architecture
• 1+ year of hands-on experience with NVIDIA NIM and Triton Inference Server for deploying optimized LLMs or specialized AI models
• Strong proficiency in CUDA C++ and CuPy for developing and accelerating custom GPU kernels and parallel algorithms
• Proven track record of translating complex theoretical papers/models into production-ready, GPU-accelerated code

Job responsibilities

• Design and Architect high-performance inference pipelines using NVIDIA NIM to serve LLMs and custom generative models at scale
• Develop and Optimize custom GPU-accelerated operators using CuPy and raw CUDA kernels to bypass CPU bottlenecks in mathematical computations
• Profile and Debug GPU memory utilization and compute kernels using NVIDIA Nsight Systems/Compute to hit aggressive latency targets
• Bridge the Gap between research-grade prototypes and production systems, ensuring code is modular, tested, and scalable
• Implement GPU-efficient data structures for real-time processing of large-scale industrial or scientific datasets
• Collaborate with cross-functional teams to integrate specialized AI microservices into broader cloud-native architectures (Kubernetes/Azure)
• Stay at the forefront of GPU computing, evaluating new NVIDIA hardware features (like H100 Transformer Engines) for project applicability

Tools

• NVIDIA: NIM, CUDA, CuPy, TensorRT, Triton Inference Server, Nsight, DCGM
• Backend/AI: Python (Expert), FastAPI, PyTorch, NumPy/SciPy, Numba
• Platform: Docker, Kubernetes, Helm, gRPC/REST, Prometheus/Grafana