Bahar Sarrafzadeh

My role broadly supports applied research within E+D by conducting original research in collaboration with various E+D product groups. My research aims to enhance, align, and evaluate LLM-powered Interactive Systems, contributing to various Copilot scenarios like Outlook, Word, and Bing.
Currently, I focus on personalizing and contextually adapting LLMs, utilizing established methods and theories from fields such as Information Retrieval, RecSys, Sociolinguistics, and AI-mediated Communication. Additionally, I leverage large-scale interaction signals to adapt these models to users’ preferences and scale the evaluation of LLM-powered Copilots.
Previously, I worked on designing, improving and evaluating Human-AI interactive systems across various domains ranging from Web search, Goal-directed browsing, email and time intelligence and writing assistance.

Before joining Microsoft, I earned my doctorate degree from the University of Waterloo (opens in new tab). My PhD research focused on Characterizing and Supporting Exploratory Search Tasks through Alternative Representations of Information (opens in new tab). During my academic journey, I engaged in fruitful collaborations with various applied research groups at LinkedIn, Google, and Microsoft. These experiences have strongly influenced my path as an applied researcher with a deep commitment to impactful research.