I’m a final-year PhD student at Stanford CS, advised by Chelsea Finn. My research is supported by OpenAI and KFAS.

My research focuses on building systems that can learn from rich textual feedback. Standard learning approaches reduce learning to optimizing a scalar “loss” or “reward”, but this single number necessarily discards useful information originally present in environmental feedback.

In many real settings, much richer feedback is available: stack traces, natural-language corrections, explanations in pairwise comparisons, or long-form reflections on failed attempts. I develop methods that leverage such signals to drive continual improvement, enabling models to refine their behavior based on such rich information.

For a technical overview, see my blog post or the selected papers below.