Responsibilities
- Design and implement low-level systems software for GPU clusters
- Work with internals of frameworks like PyTorch, NCCL, CUDA runtime—not as a user, but modifying and extending them
- Build components that make large-scale GPU training more reliable and efficient
- Debug complex distributed/concurrent systems where failures are subtle and non-deterministic
- Own systems end-to-end: from design through production
Requirements
- Systems building experience
- Designed and built complex systems—not just deployed or operated them
- Examples: Kernel subsystems, device drivers, or OS-level components
- Examples: Distributed storage, databases, or coordination systems
- Examples: Runtimes, profilers, or performance tooling
- Examples: Network stacks, protocols, or high-performance I/O systems
- Examples: Large-scale infrastructure at the systems layer
- Strong C/C++ in systems contexts (not just application code)
- Deep understanding of concurrency, memory models, and failure modes
- Experience reasoning about distributed system behavior: consistency, ordering, partial failures
- Comfortable reading and modifying large, unfamiliar codebases
Nice to Have
- GPU programming (CUDA) or GPU systems experience
- High-performance networking (RDMA, InfiniBand)
- ML framework or runtime internals
- Cluster scheduling or orchestration systems
Benefits
- Challenging projects
- A friendly and inclusive workplace culture
- Competitive compensation
- A great benefits package
- Catered lunch
- Equity program eligibility including stock options subject to approval and vesting
Additional Information
- Compensation for this position will vary based on the skills and experience you bring, as well as internal equity considerations
- In addition to cash compensation, this role is eligible to participate in the company’s equity program, which may include stock options granted in accordance with the company’s equity plan and subject to approval and applicable vesting schedules
