About the job
An applied research team within NVIDIA’s Networking Systems & Software Architecture group is solving some of AI’s hardest infrastructure problems. The team builds systems-level software that moves data between GPUs, nodes, and storage at the speed modern AI demands—spanning low-level transport optimization, hardware-software co-design, and communication frameworks that plug directly into production AI stacks. The team's charter expands into emerging domains including quantum computing interconnects. The Senior Architect role is to own modules and projects end-to-end—from scoping research questions to shipping production code. It calls for a recognized expert who drives technical decisions, pulls in ideas from research and industry, and regularly prototypes new approaches to prove a point. The work lives at the boundary of applied research and production engineering!
Responsibilities
Architecting and implementing high-performance communication and memory management libraries for distributed AI
Driving hardware-software co-optimization with GPU, DPU, NIC, and switch teams through GPUDirect RDMA, NVLink, and next-generation interconnects
Profiling and optimizing data movement across GPU memory, system DRAM, NVMe, and network fabrics
Integrating networking capabilities into AI serving stacks such as vLLM, SGLang, and TensorRT-LLM
Contributing to and maintaining open-source projects, mentoring engineers, conducting design reviews, and prototyping experimental technologies to evaluate their viability
Qualifications
Minimum
8+ years in systems software and/or networking with demonstrated ownership of complex projects.
MS, PhD or equivalent experience in Computer Science, Computer Engineering, Electrical Engineering, or a related field.
Solid understanding of high-performance networking: InfiniBand, RoCE, RDMA, NVLink, GPUDirect.
Strong C/C++/Rust systems programming with comfort in performance profiling and low-level debugging.
Understanding of ML systems concepts—transformer architectures, KV cache mechanics, model parallelism, or distributed training and inference patterns.
Preferred
Knowledge of ML inference frameworks (vLLM, SGLang, TensorRT-LLM) and their communication requirements.
Knowledge of storage networking (NVMe-oF, GPUDirect Storage, S3).
Background of Reinforcement Learning systems.