Member of Engineering (Pre-training / CUDA)

ABOUT POOLSIDEIn this decade, the world will create Artificial General Intelligence. There will only be a small number of companies who will achieve this. Their ability to stack advantages and pull ahead will define the winners. These companies will move faster than anyone else. They will attract the world's most capable talent. They will be on the forefront of applied research, engineering, infrastructure and deployment at scale. They will continue to scale their training to larger & more capable models. They will be given the right to raise large amounts of capital along their journey to enable this. They will create powerful economic engines. They will obsess over the success of their users and customers.Poolside exists to be this company: to build a world where AI will be the engine behind economically valuable work and scientific progress. We believe the fastest way to reach AGI lies in accelerating software development itself, by reshaping the developer experience with agentic systems, coding assistants, and the frontier models that power them. We deploy these systems directly into the development environments of security-conscious enterprises.ABOUT OUR TEAMWe were founded in the US and have our home there, but our team is distributed across Europe and North America. We get our fix of in-person collaboration (and croissants) in Paris each month for 3 days, always Monday-Wednesday, with an open invitation to stay the whole week. We also do longer off-sites once a year.Our team is a multidisciplinary blend of research, engineering, and business experts. What unites us is our deep care for what we build together. We’re in a race that requires hard work, intellectual curiosity, and obsession; to balance this intensity, we’ve assembled a team of low ego and kind-hearted individuals who have built the special culture Poolside has. By building collaboratively and with intention, we create a compounding effect that moves the entire company forward towards our mission: reaching AGI through intelligence systems built for software development.ABOUT THE ROLEYou would be working in our pre-training team focused on building out distributed training of Large Language Models (LLMs). This is a hands-on role that focuses on optimizing large-scale training runs via custom kernel development. You will have access to thousands of GPUs to verify changes.Strong engineering skills are a prerequisite. We assume perfect knowledge of profiling tools, CUDA, and distributed training. We look for fast learners who are prepared for a steep learning curve and are not afraid to step out of their comfort zone.YOUR MISSIONTo make our training of the best foundational models for source code generation in the world faster.RESPONSIBILITIESProfile large-scale training workloads and identify communication and computation bottlenecksCustom kernel development to improve training performanceCollaborating with researchers to make novel research ideas scale efficientlyEnhance and maintain our training and inference codebasesWrite high-quality Python (PyTorch), Cython, C/C++ code. Perform refactoringsWork in the team: plan future steps, discuss, and always stay in touchSKILLS & EXPERIENCEUnderstanding of Large Language Models (LLM)Basic knowledge of TransformersKnowledge of distributed trainingStrong CUDA background/experience with GPU programmingDevelopment experience with NCCL, CUTLASS, CUBLAS, etcUnderstanding of NVLink, NVSwitch, NVShmemStrong engineering backgroundProgramming experienceLinuxStrong algorithmic skillsPython with PyTorch, or JaxC/C++Use modern tools and are always looking to improveStrong critical thinking and ability to question code quality policies when applicablePROCESSIntro call with one of our Founding EngineersTechnical Interview(s) with one of our Founding EngineersTeam fit call with the People teamFinal interview with one of our Founding EngineersBENEFITSFully remote work & flexible hours37 days/year of vacation & holidaysHealth insurance ... (truncated, view full listing at source)