About me

I’m interested in learning from human-generated data, using ML and NLP methods in settings that postitively affect human users, and thinking about how well systems work in use-case.

Recently, my interests led me to work on NLP research projects related to text generation in dialogue systems. I pursued a variety of projects as a postdoc at UC Davis that considered dialogue systems in counseling settings, e.g., training counselors or exploring the limitations of automated support systems. During these projects I also grappled with problems generative dialogue systems face more generally, such as generic responses and uncontrollable, potentially toxic, text.

I found my way into dialogue systems from my graduate research in Computer Science at UC Berkeley. My research at Berkeley also considered applications for mental health, but used ML in digital health settings. My research included a project to infer mood passively from smartphone sensor data. The goal was to develop algorithms for a tool that could help individuals understand and monitor their wellbeing. I also contributed to statistical data analyses that explored and evaluated how well a text message program improved the participation in and efficacy of group CBT therapy for patients with clinical depression.

Throughout my projects, I’ve led multiple user studies to collect human-generated datasets of various types and to evaluate systems. These user studies (and the preceding ethics review proposals) have guided me to think extensively about the impact of systems in use and how users would interact with and, hopefully, benefit from the systems. This experience has led me to be particularly interested in how to evaluate AI methods for appropriate and reliable performance on complex tasks. The various applied projects I have contributed to have also enabled me to work on multi-disciplinary teams with wonderful collaborators, which I have greatly enjoyed.