Abstract
AI will probably be the most consequential technology of humanity. As someone who gives credence to this statement, I believe it behooves me to explore and understand AI at a technical and societal level. To this end, my technical research team used a nascent LLM architecture to build a smart search tool, allowing students to more effectively navigate the vast and complex network of academic resources at UVA. In my STS research I turned my attention to the infrastructure that enables these systems. The buildout of AI infrastructure is facing increasing pushback in American communities, with a "little us versus big them" populist movement taking shape around data center expansion. Water has become one of the central flashpoints in this resistance, but it is unclear whether water itself is the grievance or if it has become the most legible vehicle for a larger one. To better understand this nuanced dynamic, I sought to research how leading Virginia newspapers have framed water usage in their coverage of AI data center expansion. Together, the two projects engage with AI from opposite ends of the development pipeline – applying frontier techniques and investigating how their costs are framed to the people who bear them.
My capstone team built PRAGUVA (Personalizable RAG for UVA), a Graph-based Retrieval-Augmented Generation (RAG) system that helps UVA students navigate the university's academic resources. Rather than returning lists of links for students to sift through, PRAGUVA constructs a knowledge graph of UVA's professors, courses, and research outputs, then traverses it to generate direct, synthesized answers. The system distinguishes itself in two ways. First, it personalizes responses using student-specific context such as transcripts and prior interactions. Second, it is more transparent: the interface visualizes the sub-graph traversed to produce each answer, allowing users to audit the reasoning path of the output. The deliverable is a local application that grounds LLM outputs in verifiable, structured knowledge.
In my STS paper I examine how leading Virginia newspapers framed water usage in their coverage of AI data center expansion between 2023 and 2026. Using Entman's (1993) framing theory and a corpus of 12 articles from five Virginia newspapers, the paper argues that this coverage misaligns its diagnosis with its prescription. Reporters invoke water usage as a salient environmental harm but call for increased transparency rather than ecological protection, framing a legitimate governance concern as an environmental crisis and leaving readers without the context to evaluate either claim independently.
The projects barely overlapped, and that was precisely what made working on both worthwhile. Building PRAGUVA placed me inside the technical work of making AI useful whereas my STS research pulled me to the opposite end, examining AI as physical infrastructure being built in people's communities. As someone who engages with AI frequently, I often take for granted that its benefits are intuitive. They are not. For many Americans, Virginians especially, the most apparent face of AI is not a chatbot but the data center being constructed nearby, drawing on local power and water. Working on both projects reminded me that meaningfully engaging with AI means holding both vantage points at once: the view from inside, where the promise feels self-evident, and the view from outside, where the costs often arrive before the benefits do.
