We use cookies. Find out more about it here. By continuing to browse this site you are agreeing to our use of cookies.
#alert
Back to search results

Senior Software Engineer

Microsoft
United States, Washington, Redmond
Aug 20, 2025
OverviewWe are the AI Frameworks team at Microsoft, part of the CoreAI organization, and we are at the forefront of driving innovation in large-scale AI. Our mission is to enable state-of-the-art large language model (LLM) training and inference through deep optimization across the full software and hardware stack. We work across PyTorch, ONNX, and other open AI frameworks, pushing the boundaries of performance, scalability, and efficiency on a wide range of hardware accelerators. This position is part of a highly specialized sub-team focused on building the end-to-end software stack for Microsoft's first-party (1P) AI accelerators. Our work is foundational to Microsoft's long-term AI infrastructure strategy, powering next-generation models and services across Azure and Microsoft products. We collaborate closely with hardware architects, compiler teams, and model experts to co-design software solutions that unlock the full potential of custom silicon. The work is deeply technical and impactful-advancing AI systems used at global scale. If you are passionate about AI systems, low-level performance optimization, and solving hard cross-discipline engineering problems, we invite you to join us and help shape the future of AI at Microsoft.
ResponsibilitiesImplement and optimize components of the AI software stack targeting Microsoft's 1P AI accelerators.Collaborate with hardware, compiler, and model teams to develop high-performance solutions.Contribute to framework integration work for PyTorch and ONNX with custom hardware backends.Analyze performance bottlenecks and propose optimizations across framework, runtime, and hardware layers.Write clean, maintainable, and well-tested code, and participate in design/code reviews.Stay informed on emerging AI framework and accelerator technologies.
Applied = 0

(web-5cf844c5d-2fvnj)