[HN] Machine Learning System Software Engineer

- Develop an on-the-fly IR constructor that describes the semantics of an AI application as a data flow graph at runtime - Develop a runtime system to distribute computational tasks to multiple nodes in a cluster, and to virtualize computing resources (GPU/NPUs) - Integrate high-speed communication technologies (e.g., RDMA and RoCE) and low-level libraries (e.g., Verbs and UCX) into the distributed runtime system and optimize communication between AI applications and GPUs - Develop PyTorch, TensorFlow, and Triton Inference Server compatibility layers to bridge existing AI applications and our software stack - Optimize the full software stack to reduce the non-computational overhead of AI applications

Minimum Qualifications - Bachelor's degree in computer science, engineering or a related major; or equivalent practical experience - Ability to develop software in C++ and Python - Deep understanding of computer systems - Knowledge about AI frameworks such as PyTorch/TensorFlow and its internal structures Preferred qualifications - Master's, PhD degree, further education or experience in computer system, distributed computing, or a related field - 1+ years experience in developing system software for GPU/NPUs - Experience with PyTorch, TensorFlow, or Triton Inference Server, and knowledge of their internal structure - Experience in implementing neural network models from scratch - Experience in developing programs that run on a cluster system, including experience in MPI or NCCL - Knowledge of how Python programs internally work - Experience with OpenCL or CUDA - Contributions to research communities/efforts, including publishing papers at major conferences (e.g., ASPLOS, HPDC, ISCA, Micro, PACT, PLDI, PPoPP, and SC) - Ability to logically and quantitatively analyze the training process and results of AI model - Experience of more than 2 years as a machine learning engineer