Derek is an AI Research Engineer who specializes in designing intelligent and reliable AI agents through natural language understanding. His research is focused on techniques to measure ambiguity so that a dialogue system can recognize and handle uncertain or out-of-scope scenarios. He is also interested in methods to efficiently train LLMs in low resource settings including data augmentation, data denoising, and user simulators for reinforcement learning. He ultimately believes that a data-centric view of machine learning will usher in a wave of useful and trustworthy personal assistants, going far beyond just the coding agents we see today.

Derek started his research journey at the Stanford NLP Lab, working on negotiation dialogues. He eventually graduated with a masters in Computer Science from the University of Washington, while working on data collection at UW NLP. He was then advised by Prof. Zhou Yu at Columbia University studying the intersection of dialogue systems and data efficiency. He currently works as a Member of Technical Staff building AI agents for data.

This blog is about the journey of building a qualitatively superior virtual assistant and in scaling it towards a platform. Specifically, the assistant will focus on listening rather than responding, understanding rather than reacting. In other words, the assistant will move beyond executing commands and towards conversations that last more than one turn.