Abstract
Universities are increasingly using AI and data systems to manage two connected problems in undergraduate education: the practical difficulty of organizing academic work across fragmented digital platforms and the broader question of how much authority should be delegated to automated systems that guide, monitor, and evaluate students. My technical project and STS research paper approach this shared problem from different sides. The technical project focuses on academic planning. Students often reconstruct their workload by moving among Canvas pages, syllabi, calendars, and course announcements, which makes it hard to anticipate busy weeks or begin long-term assignments early. The STS project focuses on governance. As AI tools become part of advising, writing support, analytics, and integrity systems, universities are not only helping students but also redefining responsibility, merit, and acceptable conduct through platform design and policy. Taken together, the two projects ask how AI can reduce the cognitive burden of university life without making students passive objects of optimization or shifting institutional responsibility onto them. This coupling matters because a planning tool is never only a technical convenience. It also participates in decisions about what should be visible, what kinds of behavior count as responsible, and how students are expected to respond to machine-generated guidance.
My technical project, Dynamic Schedule Planner: Automated Workload Balancing via Syllabus and Canvas Extraction, addresses the planning side of this problem by building a web application that consolidates deadlines from syllabi and Canvas into a unified semester view. The system combines document parsing, a structured LLM extraction pipeline, one-way Canvas API synchronization, a workload heatmap, and a conversational interface with separate Ask and Adjust modes. The rationale was to reduce the time students spend manually reconstructing schedules and to provide earlier visibility into weeks where work clusters. Evaluation combined technical validation with structured beta testing. In syllabus workflows, all uploads produced usable calendars and more than 80 percent of assignments and dates were correctly structured, with parsing usually completing in 20 to 30 seconds. Canvas imports performed better, completing in about 4 to 10 seconds with complete correspondence to tested course pages. Five undergraduate testers completed the core workflows successfully, and usability ratings ranged from 4/5 to 5/5. Participants reported that the heatmap made heavy weeks obvious within seconds and encouraged earlier planning for projects and midterms. The project therefore shows that a unified, predictive schedule view can improve short-term planning decisions, while its limitations point to future work on OCR, complex syllabus formatting, and longer-term studies of whether better forecasting changes academic outcomes.
My STS research paper, The Algorithmic Advisor in Higher Education: How AI Systems Construct Responsibility, Merit, and Good Academic Behavior, studies the governance side of the same problem. Rather than evaluating one tool in isolation, the paper asks how AI systems in universities construct and legitimize definitions of responsible student behavior. Using documentary research and qualitative content analysis, I examined official guidance from the University of Virginia, Cornell, Columbia, George Mason, and the University of Arizona, along with platform materials from Canvas, Turnitin, Grammarly for Education, and Proctorio. The paper argues that higher education AI increasingly operates as a governance infrastructure rather than as a neutral support layer. University documents consistently frame AI use in terms of permission, disclosure, verification, and personal accountability, while vendor materials frame their systems in terms of dashboards, reports, support, monitoring, and review. Together, these sources show that AI tools make academic conduct more visible and more classifiable, even when they are marketed as helpful or transparent. The central claim is that institutions and platforms mutually shape one another: universities define the problems they want solved, such as integrity, efficiency, and retention, and platforms return those priorities in the form of categories, indicators, and recommendations that influence how students are judged. The paper concludes that algorithmic authority in higher education grows not because systems replace human judgment, but because their outputs become routine reference points for educational action.
Considered together, these projects contribute both a design response and an analytic framework for the same institutional challenge. Dynamic Schedule Planner demonstrates that AI can be used to make semester planning clearer, more anticipatory, and more manageable for students. The STS paper shows why such systems must also be evaluated in terms of power, visibility, and accountability, not just convenience or accuracy. In that sense, the projects are coupled at the level of both topic and implication: the technical work proposes one model of student support, and the STS work clarifies the social stakes that should shape how that support is built and deployed. The most important lesson from the year is that better academic AI should not only organize information more effectively. It should also preserve student agency, make system logic contestable, and avoid treating responsibility as something that can simply be transferred to a dashboard, report, or recommendation.
