Principal Machine Learning Engineer at Pinterest. Previously built AI systems at Meta (AI Glasses), Google (Gemini, NotebookLM), Amazon (Alexa), Microsoft (Cortana), and Conversica. Former faculty at Northwestern, UC Berkeley, and Mayo Clinic. 120+ publications and patents in LLMs, deep learning, and NLP.
From academic NLP research to the models behind Gemini, Alexa, and Cortana.
Technical books on machine learning and AI safety.
A comprehensive textbook that starts with probability — not loss functions — and never leaves it. Central thesis: maximum likelihood estimation and empirical risk minimization are two views of the same mathematical reality. Every chapter connects classical theory to modern AI (GPT, RLHF, diffusion models) and industry practice at the scale of hundreds of millions of users. No prerequisite beyond curiosity and basic algebra.
A practitioner's guide to building safety systems for generative AI agents deployed at consumer scale. Traditional content moderation assumes harmful content arrives as discrete artifacts from identifiable actors — but AI agents that compose user data with generative capabilities produce emergent harms invisible to any single-component review. PINSAFE introduces a seven-dimension taxonomy (Privacy, Integrity, Nonmaleficence, Security, Accountability, Fairness, Explainability) comprising 52 harm types, grounded in GDPR, OWASP LLM Top 10, and NIST AI RMF. Covers red teaming, automated evaluation, intent gating, content filtering, and monitoring.
Highlights from 120+ publications and patents.