[https://docs.google.com/presentation/d/1KB8SlKAfSqyKJ7MtuXTRtoC-ejsp11Id/edit?usp=sharing&ouid=100090609942785485941&rtpof=true&sd=true](https://docs.google.com/presentation/d/1KB8SlKAfSqyKJ7MtuXTRtoC-ejsp11Id/preview?usp=sharing&ouid=100090609942785485941&rtpof=true&sd=true)
Our research focuses on building personalized language models that can better reflect how different legal professionals reason, decide, and interact in legal contexts. We develop methods for extracting profile defining features from pairwise preference and negotiation datasets, then use those features to condition model outputs. We evaluate whether persona-aware models outperform non-personalized baselines on tasks involving preference predictions and strategic behavior. Through this work, we aim to identify the key differentiating traits that define legal preferences, accurately model and replicate individual decision-making in legal contexts, and ultimately build more personalized legal AI.
Esmie Hurd is a Research Fellow at the Stanford Law School liftlab, leading work on AI personas in the law. Her research explores LLM personalization for legal professionals and the extraction of interpretable features to explain variation in legal work product and decision making. She graduated from Yale University last May with a degree in computer science and mathematics, with a strong focus on formal logic. Her thesis explored the combined use of LLMs and Satisfiability Modulo Theories to translate legal code into First Order Logic and construct models to represent legal scenarios.
Ashley Jun is an Electrical Engineering and Computer Science student at Stanford University, focusing on the intersection of AI and legal systems with the Stanford Law School liftlab. She currently leads the development of AI-driven tools for workflow simulation in collaboration with Stanford Law School and industry partners. Her work centers on building full-stack systems and exploring the personalization of LLMs for legal context.
