Zhenwei Dai (戴振威)
Experience
Microsoft M365 Copilot
Principal Applied Scientist · Redmond, WA
December 2025 – Present
- Lead LLM post-training, including supervised fine-tuning, reinforcement learning, and on-policy distillation, to improve response quality and agent inference efficiency.
- Develop offline and online evaluation benchmarks to measure model quality and validate improvements in production.
Amazon Ads
Senior Applied Scientist · Seattle, WA
May 2025 – November 2025
- Led science strategy and the post-training roadmap for a tool-augmented LLM agent that helps advertisers set up campaigns, optimize budgets, and address business questions.
- Built a distributed post-training platform on Amazon EKS for supervised fine-tuning and reinforcement learning.
Amazon Search
Senior Applied Scientist · Palo Alto, CA
May 2023 – April 2025
- Adapted LLMs for Amazon Rufus through prompt engineering, improving response grounding and customization for product scenarios.
- Developed and deployed transformer-based query understanding models to connect customer search intent with the product catalog.
- Mentored research interns on NLP and LLM projects, leading to collaborative publications.
AWS AI Research and Education (AIRE)
Applied Scientist · Santa Clara, CA
September 2022 – April 2023
- Developed LLM-based synthetic benchmarks for evaluating the discovery of relationships between tabular datasets.
- Built machine learning algorithms for identifying joinable and unionable tables to support data discovery in AWS data lakes.
Education
Rice University — PhD
September 2018 – August 2022
- Advisors: Anshumali Shrivastava and Reinhard Heckel.
- Research: efficient large-scale machine learning, randomized algorithms, and data mining.
Publications
See the selected publications page or my Google Scholar profile for my research papers.