Abstract
The rapid rise of artificial intelligence is not just changing how we build systems, but also how we understand their impact on society. In my capstone project, I developed a centralized policy-querying system for University of Virginia Contracted Independent Organizations (CIOs) to simplify how students access and interpret complex university policies. The motivation behind this project was to reduce the inefficiencies and confusion students face when navigating scattered policy documents and to create a tool that improves accessibility and decision-making. In parallel, my STS research paper examines how generative AI is reshaping the white-collar job market, with a focus on both its intended benefits and unintended consequences. I pursued this research to better understand how AI adoption affects workers, corporate organizations, and broader labor structures. These two projects are connected through a shared focus on AI as a sociotechnical system. While my capstone explores how AI can improve user efficiency and access to information, my STS research critically evaluates how similar technologies can also redistribute power, opportunity, and risk across society.
My capstone project addresses the problem of inefficient policy navigation for student leaders by introducing a Retrieval-Augmented Generation (RAG) system that allows users to ask natural language questions and receive clear, cited answers. CIO leaders often spend significant time searching through multiple documents or emailing administrators to verify policies, creating unnecessary delays. Our system ingests relevant UVA policy documents, retrieves the most relevant passages, generates grounded responses with citations, ensuring transparency and accuracy. The methodology involved building a prototype pipeline that chunks and embeds policy documents, stores them in a vector database, and uses an LLM to generate responses constrained to retrieved sources. This system leverages various advanced RAG techniques to generate clear cut actionable solutions to queries. There is also an admin facing page where UVA Student Affairs staff can upload, delete, or replace documentation to ensure the vector database is as up to date as possible. We evaluated the system using a unit testing to validate basic functionality of the system, a predefined question bank with known answers, as well as usability testing with CIO leaders to measure time saved, clarity, and accuracy.
The results of the capstone project demonstrate that a focused, domain-specific AI system can significantly improve efficiency and user experience. Users were able to retrieve policy information faster and with greater confidence compared to manual lookup methods. The inclusion of citations increased trust and allowed users to verify responses independently. However, the project also revealed limitations, such as difficulty handling ambiguous policies and the need for clear disclaimers to avoid over-reliance on the system. Overall, the project concludes that while AI tools can streamline information access, careful design decisions are necessary to ensure transparency, usability, and ethical deployment, particularly when the system is not an authoritative source.
My STS research paper investigates the question: how is generative AI reshaping the white-collar job market? This question is significant because AI is increasingly automating cognitive tasks traditionally associated with stable, high-paying jobs, challenging assumptions about job security, and career progression. The study uses a qualitative methodology grounded in Robert Merton’s framework of manifest and latent functions to analyze both the intended benefits and unintended consequences of AI adoption. By conducting a structured literature review of economic reports, policy documents, and industry case studies, the research evaluates how AI is implemented and experienced across different institutional contexts.
The findings of the STS paper show that generative AI produces clear productivity gains and efficiency improvements, but these benefits coexist with significant unintended consequences. Evidence from sources such as the IMF and Upjohn Institute indicates that AI contributes to labor market polarization, disproportionately benefiting high-skilled workers while placing lower-skilled, routine roles at risk. Additionally, the erosion of entry-level roles suggests long-term disruptions to career pathways and skill development. The research concludes that AI is not inherently beneficial or harmful; rather, its impact depends on how it is implemented and governed. Ultimately, generative AI should be understood as a sociotechnical system that redistributes opportunity and risk, reinforcing the importance of intentional design and policy intervention.
Together, these projects highlight a central theme in my work: building AI systems requires not only technical capability but also an understanding of their broader societal implications. While my capstone demonstrates how AI can be leveraged to solve real user problems, my STS research emphasizes the importance of critically evaluating who benefits from these solutions and at what cost.
