Join Microsoft Foundry, a team dedicated to building the infrastructure that accelerates AI solutions at scale. As a Principal Applied Science Manager, you will spearhead advancements in model and agent quality through post-training techniques and significantly reduce inference latency and costs.

This role offers a unique opportunity to shape the scientific direction of cutting-edge AI technologies, mentor and develop talented scientists, and remain actively involved in experimental work and coding. You will collaborate closely with training, serving, and product teams to translate research breakthroughs into reliable, customer-facing capabilities.

Microsoft's commitment to empowering every person and organization is at the core of our mission. We foster a culture of growth mindset, encouraging continuous improvement and innovation. Become part of a team that is actively shaping the future of AI.

Lead and develop a team of applied scientists, providing coaching, setting clear goals, and delivering performance feedback to cultivate future technical leaders.

Define and own the science roadmap for post-training and inference optimization, translating customer needs into testable hypotheses and prioritizing experiments for maximum impact.

Drive advancements in supervised fine-tuning, preference optimization, and reinforcement learning, focusing on developing reward models and signals to enhance reasoning, task completion, and tool usage.

Build and manage robust data and feedback pipelines in partnership with engineering teams, ensuring clear data permissions, versioning, and separation for training and evaluation.

Establish reproducible evaluation methodologies with strong baselines and controlled comparisons to measure generalization, safety, and prevent data leakage.

Spearhead inference optimization experiments including quantization, distillation, and speculative decoding to minimize computation and memory usage, while meeting strict quality targets.

Collaborate with systems engineers on distributed training, agent rollout, batching, and caching to reduce costs and improve efficiency without compromising convergence or quality.

Transition successful experiments into production environments with staged releases, comprehensive monitoring, and rollback plans, ensuring data privacy and integrity.

Required Qualifications: - Master's Degree in Computer Science, Machine Learning, Statistics, Engineering, or a related technical field, coupled with 8+ years of relevant applied science, ML, or engineering experience. - Alternatively, a Bachelor's Degree in a related technical field with 10+ years of relevant experience, or equivalent practical experience. - Proven experience in managing applied scientists or ML researchers, including hiring, coaching, performance management, and team-based delivery. - Hands-on experience in delivering both large language model post-training and inference optimizations, demonstrating measured improvements in model quality, latency, or cost. - Strong proficiency in Python and practical experience with frameworks such as PyTorch, JAX, or comparable alternatives. - Solid understanding of machine learning, optimization, statistics, and experimental design principles. - Familiarity with distributed training and GPU serving bottlenecks, including compute, memory, and communication trade-offs. - A demonstrated track record of translating research into production capabilities through collaboration with engineering and product partners, with clear articulation of technical trade-offs.

Preferred Qualifications: - Experience with agent post-training, reward modeling, preference learning, or RL systems. - Familiarity with training toolchains like TRL and serving runtimes such as vLLM, SGLang, or TensorRT-LLM. - Experience in co-designing model improvements and serving optimizations. - Contributions to research, open-source, or production environments showcasing reproducible scientific results and practical customer impact.