Principal/Senior Applied Scientist, Security Models Training

Overview

The Security Models Training team at Microsoft is rapidly expanding to pioneer a new GenAI architecture tailored for the unique challenges of cybersecurity. This role offers a rare opportunity to contribute to frontier research while maintaining a strong product focus.

As a key member of the Security Models Training team, you will be instrumental in building and operating large-scale AI training and adaptation engines that power Microsoft Security products. Your work will transform cutting-edge research into reliable, production-ready capabilities.

As a Principal/Senior Applied Scientist, you will take ownership of the entire model development lifecycle for critical security scenarios. This includes developing novel model architectures, leading continual pre-training, fine-tuning for specific tasks, implementing reinforcement learning, and driving objective, benchmark-driven evaluation.

A significant part of your role will involve enhancing training efficiency and reliability on distributed GPU systems, deepening model reasoning and tool-use capabilities, and meticulously embedding Responsible AI, privacy, and compliance into every stage of the workflow. This is a hands-on, impact-focused position where you will collaborate closely with engineering and product teams to translate innovations into measurable, shipped outcomes, define quality gates, and mentor other scientists and engineers to scale results across globally distributed teams.

You will leverage strong coding, experimentation, and debugging skills with a systems mindset to accelerate iteration cycles, improve throughput and cost-effectiveness, and help shape the next generation of secure, trustworthy AI solutions for Microsoft's customers.

Responsibilities

• Technical Leadership & Ownership: Set the technical direction for major security domain initiatives and lead security model programs encompassing pre-training, task tuning, reinforcement learning, and evaluation, translating advanced research into production-ready capabilities.
• Advanced Model Design: Build and customize deep learning model architectures (e.g., modifying transformer blocks, attention/memory modules) at the SLM/LLM scale, making principled architectural tradeoffs to enhance reliability, robustness, and security-specific behavior.
• Advanced Model Training: Apply deep expertise in pre-training, post-training, and reinforcement learning (RL) for both language and other modalities, including time-series data.
• Design & Evaluate Datasets: Construct high-quality datasets and benchmarks, define objective evaluation frameworks and quality gates, and conduct ablation studies to measure impact and optimize data and training effectiveness for confident product decisions.
• Develop Data Infrastructure: Create and maintain scalable pipelines for the ingestion, preprocessing, filtering, and annotation of large, complex datasets, with meticulous attention to privacy, governance, and long-term reuse across various security scenarios.
• Research & Innovation: Collaborate with cross-functional teams to push both research and product boundaries, delivering models that achieve real-world impact.

Qualifications

• M.Sc. / Ph.D. in Computer Science, Information Systems, Electrical or Computer Engineering, or Data Science (Ph.D. strongly preferred).
• Candidates with M.Sc. / Ph.D. in related fields with proven industry experience or a strong publication record in LLM, Information Retrieval, Machine Learning, Natural Language Processing, Time Series Forecasting, and Deep Learning will also be considered.
• Minimum of 5 years of proven hands-on experience (including post-grad work) in building and deploying Machine Learning products.
• Key expertise areas include Natural Language Processing and Large Language Models, along with an understanding of Privacy and Responsible AI concepts.
• Demonstrated strong history of successfully translating applied research into production-ready solutions and a proven track record of delivering projects within large-scale production environments.
• Proven expertise in the LLM and/or time-series forecasting domain, demonstrating comprehensive knowledge of relevant concepts.
• Proficiency in LLM's pre and post-training, including CPT, SFT, and RL, LLM benchmarking, agentic flows, and model alignment.
• Hands-on experience in building neural model architectures at the 100M+ scale and the proficiency to adapt them at all abstraction levels, down to individual blocks (e.g., changing attention block inner workings, introducing new blocks, or changing routings).
• Demonstrated proficiency in problem-solving and data analysis, with substantial expertise in evaluating the performance of large language models (LLMs) and/or time-series forecasting models, developing benchmarks tailored to practical scenarios.

Preferred Qualifications

• Ph.D. degree in Computer Science, Information Systems, or Data Science.
• Evidence of research contributions through publications or records of top-tier journal and conference publications or submitted/accepted papers in top venues (KDD, ICML, AAAI, ACL, ICLR, etc.).
• Proven track record in pre/post-training of large transformer models for language and/or time series tasks.
• Customer obsession and a passion for making real-world product impact through production-deployed systems.
• Excellent verbal and written communication skills, with the ability to simplify and explain complex ideas.
• Effective collaboration skills while working within a globally distributed organization.

Key skills/competency

• Generative AI
• Cybersecurity
• Large Language Models (LLM)
• Deep Learning
• Natural Language Processing
• Reinforcement Learning
• Distributed Systems
• Model Architecture
• Responsible AI
• Time Series Forecasting