From messy data
to useful products.

Place records contain names in many languages, often mixed with promotional text and other clutter. The existing generative approach cost $70,000 per run and could introduce words that were never in the source name.

I trained a model to decide which parts of the original name to keep or remove. This made the task more focused and constrained the output to text already in the record.

Finding a respirator that fits can involve considerable trial and error. MasqFit investigates whether facial measurements and fit-test evidence can help people choose better-fitting respirators.

I recruited participants, collected measurements and fit-test outcomes, developed ML algorithms, and built and released an open-source iOS app. My role spanned the research question, the evidence needed to study it, and the application people can use.

Understanding the Link: The Correlation Between Student Mindsets and Chronic Absenteeism connects attendance data with student survey responses to help school and district leaders understand patterns in absenteeism.

The study’s national attendance context spans 7,000+ schools and over 11 million student-year records. Its findings connect school safety, climate, relationships, and SEL topics with attendance patterns.