Alex Deng

Alex Shaojie Deng

I am heading the engineering and science for Responsible AI in Microsoft’s Core AI division. Our platform safeguards hundreds of billions of generative AI interactions annually, providing foundational safety systems for nearly all Azure OpenAI API traffic. This includes multimodal content moderation, jailbreak prevention, cross-domain prompt injection protection (XPIA), hallucination detection, and protected material identification. We are also at the forefront of safety evaluation, red teaming, model steering, and establishing the trust and security foundations for agentic AI systems.

From 2020 to July 2024, I served as Technical Director of Platform Data Science and Machine Learning at Airbnb, where I chaired the science track of the company-wide AI & ML Council. My work spanned modernizing applied research workflows, developing a collaborative ML prototyping ecosystem, building foundational data infrastructure for event stream analytics and behavioral modeling, and advancing agile experimentation frameworks for offline and online model evaluation. I also held leadership roles in both the Relevance & Personalization team and the Core Data Science organization.

Earlier in my career, I was part of Microsoft’s Cloud & AI Platform, contributing to both the Microsoft Analysis and Experimentation Team and Azure Machine Learning. At the former, I led methodological and engineering improvements to one of the industry’s largest online experimentation platforms, with work spanning causal inference, Bayesian optimization, reinforcement learning, AutoML, recommender systems, high-throughput distributed systems, and analytics infrastructure. In Azure ML, I focused on the end-to-end experience for internal ML practitioners, including distributed training, automated ML pipelines, and product analytics.

I earned my Ph.D. in Statistics from Stanford University in 2010, where I studied sequential Monte Carlo methods under Prof. Tze Lai

Selected Publications (newest first)

Preprints

Other Talks/Slides

Other (non-peer-reviewed)