The Soft White Underbelly interview archive is hundreds of hours of unstructured first-person testimony about homelessness, addiction, and mental health. As an NLP Research Analyst at Seattle Children’s Research Institute, I built the full pipeline to turn it into structured insight: transcript extraction from the source videos, preprocessing, embeddings, then k-means clustering and Latent Dirichlet Allocation to identify latent patterns across 900+ transcripts.
A companion sentiment-analysis pipeline (NLTK, TextBlob) quantified emotional responses across mental-health surveys, enabling data-driven identification of emotional triggers.
Presented the methodology and findings at the University of Washington Research Symposium to technical and non-technical audiences.