This project was developed in collaboration with LAS, a mission-driven research center at NC State that brings together government, industry, and academic experts to support intelligence analysis. LAS works with many kinds of analysts — from language analysts to cyber/computer-network analysts and broader intelligence analysts — each with different daily workflows and analytic needs. Our goal was to design an explainable model-selection tool to help those analysts pick the most effective speech-to-text model for their audio data. To build it, we adopted a three-level explainability framework, conducted user journeys, and distributed a survey to better understand the diverse roles and preferences of different types of analysts. The aim is to support decision-making without causing cognitive overload and to give flexibility that adapts to each analyst’s unique workflow. We are now in the final stages of delivering our work to stakeholders, and the full case study will be available on my portfolio next year.