Principal Architect, AI Inference Networking - NIXL and Dynamo
Core
Architecting and driving adoption of NIXL (NVIDIA's open source Inference Xfer Library) and Dynamo to solve high-performance tensor movement in disaggregated AI inference clusters, specifically for demanding deployments in China.
Role type
Senior IC Principal Architect (AI Inference Networking)
Builds
NIXL backends, plugins, and dynamic data path APIs for elastic inference clusters.
Domain
AI Infrastructure / High Performance Networking / Distributed Systems
Deliverable
production ML models
Required skills
C++ systems programming, Python, RDMA networking (InfiniBand/RoCE), GPU memory movement, LLM serving stacks (Dynamo, TensorRT-LLM, vLLM, SGLang), distributed systems optimization, kernel/driver level code reading.
Preferred skills
Open source contributions to NIXL/Dynamo/NCCL/vLLM, experience with thousands of GPU clusters, CUDA kernel development.
Technologies
NIXL, Dynamo, InfiniBand, RoCE, CUDA, C++, Python, TensorRT-LLM, vLLM, SGLang, Triton, GPUDirect.
Responsibilities
Drive NIXL adoption from architecture to production rollout in China; profile customer inference stacks to fix throughput and TTFT bottlenecks; write NIXL backends and extend Dynamo integrations; define dynamic APIs for elastic scaling and runtime rerouting; bring regional requirements into NVIDIA's roadmap; grow the local NIXL community.
Seniority
Principal, hands-on IC with technical authority.
