ML Research Engineer

Microsoft•Redmond, WA

1d•$102,100 - $219,200

About The Position

Microsoft is hiring an ML Research Engineer to join the Systems Planning and Architecture (SPARC) team within its Azure Hardware Systems and Infrastructure organization. The team develops infrastructure and foundational technologies supporting Microsoft’s cloud businesses and services. The role focuses on designing and developing quantization and numerics algorithms for efficient deployment of large language model (LLM) inference and training in Azure production environments. The engineer will develop and evaluate model optimization tooling, investigate performance bottlenecks, prototype low-precision data formats on hardware accelerator SDKs, and collaborate with AI/ML, hardware architecture, and software framework teams to improve end-to-end system efficiency.

Requirements

Bachelor’s degree in Statistics, Econometrics, Computer Science, Electrical or Computer Engineering, or a related field and 2+ years of related experience; or a master’s degree in one of these fields and 1+ year of related experience; or a doctorate in one of these fields; or equivalent experience.
Ability to meet Microsoft, customer, and/or government security screening requirements, including passing the Microsoft Cloud Background Check upon hire or transfer and every two years thereafter.

Nice To Haves

Bachelor’s degree in a relevant field and 5+ years of related experience; or a master’s degree and 3+ years of related experience; or a doctorate and 1+ year of related experience; or equivalent experience.
At least 1 year of experience creating publications, such as patents or peer-reviewed academic papers.
Demonstrated experience delivering production-grade software in areas such as model compression, low-precision numerics (including FP8, INT8/4, NVFP4, or MX formats), low-level kernel development, and performance optimization.
Proficiency with modern deep learning frameworks, including PyTorch, TensorFlow, TensorRT, and ONNX Runtime.
Expertise in GPU/NPU kernel development using CUDA, Triton, ROCm, or comparable frameworks, and fast model bring-up on a new stack.
Strong understanding of Transformer and LLM architectures, with hands-on experience in optimization techniques such as quantization, pruning, tensor or parameter sharding, model parallelism, KV-cache optimization, and Flash Attention.
Practical experience with large-scale model evaluation, including benchmarking state-of-the-art LLMs and fine-tuning large models using SFT or RL.
Strong programming skills in Python, C, and C++.
Excellent communication skills and ability to collaborate effectively in hybrid, team-oriented environments.
Hands-on experience implementing and optimizing low-level linear algebra routines, including custom BLAS kernels.
Deep knowledge of mixed-precision arithmetic units, including numerical formats and microarchitecture.

Responsibilities

Design and develop novel quantization and numerics algorithms for efficient LLM inference and training in Azure production environments.
Drive software development and model optimization tooling proof-of-concept efforts to streamline deployment of quantized models.
Analyze performance bottlenecks in quantized state-of-the-art LLM architectures and drive performance improvements.
Prototype and evaluate emerging low-precision data formats through proof-of-concept implementations on novel hardware accelerator SDKs.
Co-design model architectures optimized for low-precision deployment in collaboration with companywide AI/ML teams.
Work cross-functionally with data scientists and ML researchers and engineers to align on model accuracy and performance goals.
Partner with hardware architecture and AI software framework teams to ensure end-to-end system efficiency.