Principal Applied Scientist

** Overview*
* Frontier Tuning is Microsoft's AI customization platform that enables enterprises to adapt foundation models to their unique workflows, domains, and data-while preserving security, privacy, and reliability  we grow, Frontier Tuning is becoming a critical pillar for ensuring enterprise‑specific capabilities are systematically learned and reflected across Microsoft 365 and beyond.

We are seeking a  
** Principal Applied Scientist
** with strong research and systems‑building skills who is excited to push the frontier of large‑scale model post‑training and adaptation. This role spans algorithmic innovation as well as the design and development of scalable infrastructure and tooling for training, steering, evaluating, and securely deploying enterprise‑ready AI systems.

Post‑training may include reinforcement learning, fine‑tuning, architectural modification, inference‑time control, evaluation‑driven adaptation, or privacy‑preserving training techniques applied under real‑world enterprise deployment constraints.

Ideally, candidates will have experience in one or more of the following areas:

+ Scalable training systems for RLHF/RLAIF or other post‑training pipelines.

+ Scalable inference systems for LLMs.

+ Transformer architecture design or efficient adaptation techniques (e.g., LoRA-style methods).

+ Inference‑time steering, cont rollability, or alignment approaches.

+ Privacy-preserving machine learning (e.g., differential privacy or secure training).

+ Debugging, evaluation, or development tooling for foundation models.

+ Multimodal model training, including language, vision, or diffusion models.

This position is based at the Redmond campus with 3 days per week work in the office and 2 days per week work from home. Domestic relocation assistance is available.

Microsoft's mission is to empower every person and every organization on the planet to achieve more. As employees we come together with a growth mindset, innovate to empower others, and collaborate to realize our shared goals. Each day we build on our values of respect, integrity, and accountability to create a culture of inclusion where everyone can thrive at work and beyond.

** Responsibilities*
* + Design and develop methods to adapt foundation models (e.g., language, diffusion, or multimodal models) for enterprise‑specific tasks such as document understanding, workflow automation, or content generation.

+ Contribute to one or more aspects of the post-training stack, including:

+ Scalability and efficiency of training and inference systems

+ Reinforcement learning or fine-tuning methods

+ Architectural or parameter‑efficient adaptation techniques

+ Inference‑time steering or cont rollability approaches

+ Tooling for evaluation, debugging, or model development

+ Privacy- or security‑preserving training techniques (e.g., differential privacy)

+ Harnesses

+ Implement and evaluate adaptation approaches under real‑world enterprise deployment constraints such as latency, safety, privacy, policy compliance, and compute efficiency.

+ Partner with research and engineering teams to translate product or customer requirements into scalable model adaptation solutions.

+ Explore post‑training techniques that improve domain specialization, tool use, planning, or agentic behaviors in enterprise environments.

+ Drive technical work from concept to prototype, delivering new methods, systems components, or empirical insights that advance enterprise model customization.

+ Document approaches and share best practices to improve organizational capabilities in post‑training and secure deployment of foundation models.

+ Support mentorship and onboarding of interns or early‑career team members as appropriate.

** Qualifications*
* *
* Required Qualifications:

*
* +…