Software Engineer, ML InfrastructureEngineering · Full-time · San Francisco; New YorkOur mission is to automate coding. The first step in our journey is to build the best tool for professional programmers, using a combination of inventive research, design, and engineering. Our organization is very flat, and our team is small and talent dense.
We particularly like people who are truth-seeking, passionate, and creative. We enjoy spirited debate, crazy ideas, and shipping code.About the RoleThe ML Infrastructure team builds large-scale compute, storage, and software infrastructure to support Cursor's work building the world's best agentic coding model. We're looking for strong engineers who are interested in building high-performance infrastructure and the software to support it.
This role works closely with ML researchers and engineers to enable their work through improvements to our training framework, systems reliability/performance, and developer experience.What You'll DoCollaborate with ML researchers to improve the throughput and reliability of trainingWork with OEMs, cloud service providers, and others to plan and build cutting-edge GPU infrastructureImprove the density and scalability of compute environments to enable increasingly large RL workloadsCreate software and systems to automate building, monitoring, and running GPU clustersBuild workload scheduling and data movement systems to support Cursor's growing training footprintYou May Be a Fit IfA strong background in systems and infrastructure-focused software engineering, particularly in Python, Typescript, Rust, and GolangExperience with distributed storage and networking infrastructure, particularly on Linux systems across cloud and bare metal environmentsExposure to large-scale systems and their unique challenges, ideally across thousands of nodes with significant resource footprints.Production use of infrastructure-as-code and configuration management, across hosts and KubernetesNice to HaveOperational exposure to Nvidia GPUs with Infiniband or RoCE, particularly with Blackwell and Hopper-class hardwareExposure to Ray, Slurm, or other common compute and runtime schedulers