Selected Projects

A selection of projects that show how I design behavioral measures, evaluate model performance against human judgment, and translate messy participant-facing data into clear conclusions.

Human-referenced model evaluation

Human Speech Emotion Classification

Erdős Institute data science project comparing machine-learning approaches to human performance on speech emotion classification.

  • Worked with a speech-emotion dataset to evaluate how well models classified affective categories from audio.
  • Compared support vector machine and convolutional neural network approaches against a human benchmark.
  • Examined category-level error patterns to understand where model behavior aligned with, or diverged from, human judgments.
  • Framed model performance as more than an aggregate accuracy score: the goal was to identify what kinds of emotional judgments were difficult, ambiguous, or systematically confused.

Cognitive measurement

Theory of Mind + Inhibitory Control

A research program operationalizing abstract social-cognitive judgments into measurable behavioral outcomes across development.

  • Designed and analyzed tasks measuring how children reason about others’ false beliefs while managing competing reality-based responses.
  • Used dense longitudinal data and computational modeling to study how inhibitory control contributes to change over time.
  • Translated theoretical constructs—belief representation, response conflict, developmental variability—into testable predictions and interpretable measures.
  • Relevant proceedings: CogSci 2020 and CogSci 2026.

Participant-facing research operations

Lookit + Liberty Science Museum Studies

Remote and public-facing developmental studies with children and families, where data quality depends on clear tasks, good instructions, and careful moderation.

  • Supported child-friendly study design, participant comprehension, and smooth data collection in both online and museum-based contexts.
  • Attended to task failure modes such as misunderstanding instructions, attention shifts, parent interference, and incomplete responses.
  • Helped translate live research constraints into practical decisions about materials, moderation, and data interpretation.
  • Built experience communicating research goals clearly to families and adapting procedures in public-facing settings.

How I work

Across these projects, I’m interested in the same core problem: how to make judgments observable. Whether the judgment belongs to a child, an adult participant, or a model, I care about designing measures that reveal not only whether an answer is correct, but what kind of reasoning or failure may have produced it.