Senior Principal Software Engineer

Key Skills

AWS, Azure, C++, CUDA kernels, CUDA/GPU programming, Elasticsearch, Google Cloud Platform (GCP), Kubernetes-K8s, Leadership Skill, Linux Operating system, Machine learning techniques, Python Programming, SQL

Job Description

As a distributed training software engineer, you'll build and manage large GPU clusters and help optimize model architectures for maximum training performance.

UiPath is building large multi-modal models focused on taking action and moving us closer to useful general intelligence. Our iteration speed crucially depends on training performance. In this role, you'll directly work with research and applied scientists to ensure model architectures are GPU sympathetic and optimized for maximum training performance.

This position entails building and managing large-scale complex infrastructure, requiring strong quantitative intuition and programming skills. Beyond that, you need to be kind, courteous, motivated, self-starting, honest, and responsible.

Background:

This role is with UiPath advanced machine learning research group. We focus on building large multi-modal models that can take action and move us closer to useful general intelligence. We are machine learning researchers and engineers working to solve the hard problem of getting machines to use software like humans do.

What you'll do at UiPath

• Build, manage, and maintain large GPU clusters in the cloud for training large models.
• Optimize and improve the training utilization and reliability of those clusters.
• Work directly with researchers to align model architecture with training performance.
• Stay on top of the latest research and technology and communicate your knowledge throughout the company.
• Mentor other members of the team and help hire new people.

What you'll bring to the team

• 6+ years of experience as a software engineer
• Expert understanding of distributed training concepts and tools, for example NCCL, torch.distributed, MPI, etc.
• Experience with Kubernetes and cloud providers (Google Cloud Platform, Azure)
• Experience with GPUs, interconnect (intra-node and inter-node), cluster hardware, and GPU performance optimization.
• Familiarity with Rust and/or a willingness to learn new programming languages.
• Experience with low-level CUDA programming in Rust / C++ is nice to have.
• Knowledge and appreciation of statically typed languages and using type systems to catch bugs early, rather than at runtime.
• Solid knowledge of basic data structures (list, vector, stack, heap), basic algorithms (sort, search, etc.), and associated time/memory asymptotic complexity.
• Good knowledge of GNU/Linux.

Maybe you don’t tick all the boxes above— but still think you’d be great for the job? Go ahead, apply anyway. Because we know that experience comes in all shapes and sizes—and passion can’t be learned.

Many of our roles allow for flexibility in when and where work gets done. Depending on the needs of the business and the role, the number of hybrid, office-based and remote workers will vary from team to team. Applications are assessed on a rolling basis and there is no fixed deadline for this requisition. The application window may change depending on the volume of applications received or may close immediately if a qualified candidate is selected.