My work on Gemini has focused on improving the model quality, specifically for Gemini Extensions, which enables Gemini to carry out actions on the user’s behalf.

A large part of the work is creating and managing high quality Supervised Fine Tuning (SFT) and Reinforcement Learning though Human Feedback (RLHF) training and evaluation datasets for Large Language Models (LLMs). Part of keeping the quality involves helping write, maintain, and disseminate guidelines for good conversational experiences.

I also partner with User Experience Researchers (UXR) to create user- or vendor-facing surveys to assess conversational quality in a way that is reliable, accurate, and valid.