Large language models. Gemini agents, reasoning, and alignment. Led code- RL(H)F/post-training for Bard+PaLM 2.- Areas: reinforcement learning, code, tool use agents, safety, real-world/multimodal evals- Projects: Gemini, Bard, PaLM 2- Additional launches: Cloud (DuetAI), Search (CodeTips), Colab, Android Studiobot

Dialogue generation, memory models that blend retrieval and generation.

Developed first end-to-end multi-document, abstractive summarization model. Paper accepted to ICML.

Worked on ad moderation, with a focus on the human-AI system. • Led multi-office (Menlo Park, Austin, Hyderabad) project running A/B tests on how to present model predictions and uncertainty to human reviewers.

Analyzed community and interest groups, focused on health and emergent behavior.• Built a classifier for detecting organic marketplace groups, and helped inform the creation of Facebook Marketplace in Feb 2015.• Helped set team direction to focus on smaller-sized groups in international markets.