The Summit Showcase
The Showcase will highlight exemplars in the use of AI in education, as well as research related to the use of AI in education, in a poster and demo session. During the Showcase, Summit attendees will move from station to station, engaging with the posters and demonstrations in small informal conversations and discussions.
We invited any member of the Duke community to submit a Showcase topic, including faculty, students, trainees, and staff. Submissions had the option to propose a poster, demonstration, or combined format.
Drowning in AI Video: Navigation Strategies Treading Water In A Sea Of Junk
Stephen Toback, Media Architect and Senior Producer for Academic Media Production, Duke OIT
New AI video models and features surfacing daily, the “visual truth” in academic media is shifting faster than we can document it. This showcase explores how the Duke Digital Media Community (DDMC) serves as a living laboratory and permanent record for testing these relentless updates in real-time. By benchmarking tools like Sora, Luma, and Kling against strict academic standards, we move from being “swamped” by the tech to building a collaborative framework of trust. Learn how we use community-driven research to verify what is factually useful and what is merely a “pixel dream.”
AI-Supported Simulation for Real-Time Feedback on Motivational Interviewing and Patient Interaction Skills
Diana Rodriguez Manrique, Graduate Student, Duke University School of Nursing
Today’s healthcare students have come of age in an environment where immediate feedback is the norm, largely shaped by social media and digital platforms. At the same time, the COVID-19 pandemic disrupted opportunities for in-person learning and may have contributed to challenges in developing interpersonal and therapeutic communication skills. Because effective therapeutic communication is essential across all healthcare professions, opportunities to practice these skills are critical, yet many students experience patient interactions as stressful, vulnerable, or high stakes. An AI-supported, hyper-realistic simulation platform offers learners the opportunity to engage with a lifelike avatar presenting with an authentic health concern, allowing them to practice skills such as open-ended questioning, reflective listening, and therapeutic responses in a psychologically safer environment. Following the interaction, students receive detailed feedback on communication strengths, areas for growth, the soundness of their clinical advice, and the ways in which their tone and communication style influenced the avatar’s responses, helping reduce anxiety and build confidence for real-world patient care.
Seeing Race: An App for Visualizing Human Genomic Variation Concepts
Nadeesha Perera, Instructor in the Department of Biology, Trinity College of the Arts & Sciences
This model of commonly misunderstood concepts in the study of human population genomic variation presents simulated, reduced-dimension genome-wide SNP data from a hypothetical multiethnic population. Each person has fictionally been asked to fill out four different census forms from different nations and time periods. App users can toggle among census categorizations and observe how the same individual “changes race”. Moveable race boundaries and live updating of allele and haplotype frequency variation within and among race groups provides a visual, manipulable representation of how humans vary along genomes and within and across populations. In Biology, this model’s pedagogical utility includes depicting the impact of evolutionary forces such as migration, selection, mutation and drift on human genomic variation. Broader utility may extend to educating the general public about the biological support, or lack thereof, for race categorizations.
Evaluating Engineering Students’ Detection of Hallucinations in Large Language Model Solutions
Zachary Deutsch, Undergraduate Student, and advisor Siobhan Oca, Director of Master’s Studies, Assistant Professor of the Practice in the Thomas Lord Department of MEMS, Pratt School of Engineering
Large language models (LLMs) such as ChatGPT are widely used by engineering students to complete coursework, but current literature lacks empirical evidence on whether engineering students can recognize when LLMs generate confident but incorrect outputs (hallucinations). While LLMs can provide explanations, guidance, and feedback, they can also provide incorrect solutions, raising concerns for engineering education. This study addresses this gap through a sequential design consisting of (1) measurement of GPT-5.1 hallucination rates and illustrative examples using undergraduate mechanics and electricity and magnetism (E&M) problems, and (2) a survey of engineering students’ ability to detect these hallucinations in LLM-generated solutions. To our knowledge, this is the first study focused specifically on engineering students’ detection and confidence in LLM hallucinations in problem solving.
Brain-First AI Use: Balancing AI Assistance and Student Thinking
Cheryl Beierschmitt, Peer Education Manager, Academic Resource Center
How can students use AI tools without letting AI do the thinking for them? Duke University’s Academic Resource Center developed an AI Toolkit that applies a “brain-first” framework to guide generative AI use in learning. This approach helps students use AI to support rather than shortcut their learning. We will share example prompts for common learning challenges and consider how AI can reinforce essential cognitive work such as retrieval, interleaving, and metacognition. The examples are adaptable across diverse institutional contexts and disciplines.
Integrating AI into Manufacturing Education: A Student-Driven Approach with Industry Insight
Shana McAlexander, Assistant Research Professor in the Thomas Lord Department of Mechanical Engineering and Materials Science, Pratt School of Engineering
This new course from Mechanical Engineering and Materials Science examines how artificial intelligence is transforming modern manufacturing through a blend of student-led projects and expert-led seminars. Students explore core topics including machine learning, optimization, robotics, digital twins, quality control, supply chains, ethics, and environmental sustainability. The course emphasizes active learning, with participants driving discussions, presenting case studies, and shaping project directions. Special seminars were provided from industry leaders and academic researchers providing real-world perspectives on data-driven design and smart manufacturing practices. A student-led project integrates concepts across the semester, enabling students to apply AI techniques to complex manufacturing challenges.
Designing an After‑Hours AI Chatbot for the Duke Medical Center Library
Sarah Cantrell, Associate Director for Research & Education at the Medical Center Library & Archives, and Hope Riffee, Digital Projects Analyst at the Medical Center Library & Archives
Artificial intelligence and large language models are rapidly reshaping the ways health professionals and librarians engage with information. AI tools are increasingly seen as helpful assistants that can streamline workflows and extend services. With this in mind, we identified chat reference as a service in our library that could take advantage of such AI tools. Our objective was to create an AI after hours chatbot using Duke MyGPT Builder to extend existing chat reference services to support the needs of our community of students, faculty, staff, and clinicians. In this poster, we will share both considerations for planning AI chat services as well as tips on creating a custom chatbot.
Change and Stability in Generative AI Habits Among Duke Graduate Students, Faculty, and Staff
Courtnea Rainey, Assistant Dean for Assessment and Evaluation, Graduate School, and Leo Biggs, Senior Program Coordinator, Graduate School
In 2025, The Graduate School launched the Spotlight Project, an assessment and data workflow to collect, process, and report timely survey results. As part of this Project, a survey, “Generative AI Use Among The Graduate School Community” was administered in April 2025 and April 2026.
Select findings from the 2025 survey regarding the prevalence, frequency, and contexts of generative AI use have been previously summarized. (Duke community members may access this report at https://duke.is/Spotlight-2026.) Here, we compare The Graduate School community’s current patterns of generative AI use (2026) to those from last year.
Agentic AI Personal Tutor for Plastic Surgery Education
Raunak Goyal, Research Fellow, Department of Surgery, School of Medicine, Andy Shar, MS1, School of Medicine, and Matthew Simpson
Plastic Surgery Personal Tutor is an agentic AI educational application built on a graph-based retrieval-augmented generation (Graph RAG) architecture that indexes curated plastic surgery literature into a structured knowledge graph of over 112,000 entities and 62,000 relationships. The platform integrates agentic features—training-level-adaptive learning and automated board-style question generation—with multimodal text and voice interaction. In a pilot study with plastic surgery residents (PGY1–PGY6) across 50 evaluated interactions, the application achieved a median accuracy of 5/6, completeness of 3/3, and safety of 3/3, with 74% of responses rated as safe to act upon. Automated NLP evaluation against standardized textbook references yielded a mean BERTScore recall of 0.843 and mean concept coverage of 80.7%. An agentic AI tutor grounded in verified plastic surgery references demonstrated high accuracy, completeness, and safety as rated by residents across training levels. These findings support AI-powered tutoring as a promising supplement to plastic surgery education and offers a transferable framework for designing, evaluating, and responsibly deploying Graph RAG–based AI tools in specialty medical education.
Red Teaming AI in Nursing Education: What Happens When You Stress-Test the System Students and Faculty Actually Use?
Jennie De Gagne, Clinical Professor, School of Nursing
This study applied a red team audit methodology to Microsoft 365 Copilot Chat, systematically probing the system across 31 test cases that mirror real faculty tasks in nursing education: evaluating admissions materials, verifying clinical knowledge, and assessing student work for academic integrity. Every test case revealed at least one issue, including ESL-penalized admissions scoring, clinical guidance presented without verifiable citations, and a critical-severity detection error in which the system identified human writing as AI-generated while passing AI-generated text as human. The findings expose a trust paradox: the system performed most fluently precisely where accuracy and fairness mattered most, creating risks that are difficult to detect without deliberate audit. The poster presents the three-pillar testing framework, key failure patterns, and a multi-level set of recommendations for faculty, developers, and institutional governance.
Adapting the Duke AI Ethics Learning Toolkit for Clinical Prioritization in Nursing Education
Jennie De Gagne, Clinical Professor, School of Nursing, and presented by Gina Rose, Instructor, Medical Center in the School of Nursing
Clinical prioritization requires distributive justice and moral discernment, yet nursing-specific resources for addressing ethical risks in AI-mediated decision making remain limited. This study adapted the Duke AI Ethics Learning Toolkit — an interdisciplinary, open-access active learning resource organized around guided inquiry, conversation prompts, and applied activities — into a nursing-focused mini-toolkit for generative AI use in clinical prioritization pedagogy. Three core topics from the original toolkit (“Can We Trust AI?”, “Is AI Biased?”, and “Who Benefits from AI?”) were reframed for nursing contexts and extended into multipatient case vignettes, reflection prompts, and rubric criteria anchored in three complementary ethical frameworks at the learner, professional, and institutional levels. The poster presents the adaptation methodology, demonstrates how the Duke toolkit’s inquiry-based design translates into prioritization scenarios involving social determinants and stigmatized conditions, and discusses implications for faculty development and simulation integration.
Histy: A History Reading and Annotation Environment for Externalizing Student Thinking
Yunxi Kong, Graduate Student, Masters of Economics & Computation Program, Economics
Histy is a web-based history reading and annotation environment designed to help students make their thinking visible while reading historical texts. The system combines standard reading, immersion reading with historically coherent visual context, line-anchored highlighting, note-taking, and threaded discussion. Instead of using AI only to generate content, Histy explores how AI-supported interfaces can scaffold interpretation, evidence use, and peer dialogue in history learning. The showcase will present the prototype design, learner workflow, and planned comparison between standard and immersion reading environments.
Pressing Prompts: Bringing Critical AI Conversations into the Classroom
Hannah Rozear, Librarian for Biological Sciences, Global Health, and Artificial Intelligence Learning, and Remi Kalir, Associate Director of Faculty Development and Applied Research, Duke Center for Teaching and Learning
Pressing Prompts (formerly the AI Ethics Learning Toolkit) is an openly-accessible educational resource designed to support student-centered, critical engagement with AI across disciplines. Originally developed by Duke undergraduate students, faculty, and staff, Pressing Prompts currently explores eleven questions such as “Can we trust AI?” “Is AI theft?” “Is AI a friend?” with adaptable classroom activities that include things like conversation starters and disciplinary extensions. Pressing Prompts now includes dozens of new assignments and activities that invite students to reflect on AI’s social, ethical, environmental, and personal impacts. In this showcase, the project team from Duke’s Center for Teaching and Learning and Duke Libraries will highlight recent updates, demonstrate how these materials can be used in both face-to-face and online course contexts, and share ways for Duke instructors to get involved with the project during the Fall 2026 semester.
Be You, Be Bold, Choose Duke Powered by AI, Live look at Duke’s Job Description Agent
Robert Williams, Senior Recruiter, Human Resources
This showcase presentation features a Senior Recruiter who developed an AI-powered conversion tool that transforms traditional job descriptions into mobile-friendly, branded postings aligned with Duke’s employer identity. The solution converts foundational job description language into structured, candidate-centric sections such as “Be You,” “Be Bold,” and “Choose Duke,” ensuring consistency in voice and presentation across roles. Designed with a mobile-first approach, the tool enhances accessibility and engagement for candidates applying across devices. Attendees will gain insight into the design, testing, and implementation of the AI agent, demonstrating how it improves efficiency while standardizing content at scale. Key lessons learned—including real-world application, performance insights, and workflow integration—highlight how AI can modernize job description development and elevate recruiting outcomes.
Pairthink: AI-Assisted Active Learning in Remote or In-person Groups
Eric Green, Director of Undergraduate Studies, Duke Global Health Institute; Associate Professor of the Practice of Global Health
Pairthink is a web platform that supports active learning across in-person, remote-synchronous, and remote-asynchronous courses. Instructors author an activity once — a sequence of questions with rubrics, figures, and uploaded readings — and choose how to deploy it: as a live class session where students co-author answers in a shared, sub-200ms collaborative editor with peer typing indicators, or as a solo activity students complete on their own schedule. Every AI surface (submission feedback, adaptive hints, and post-class instructor insights) is grounded in the instructor’s own rubric, framing, and readings via retrieval-augmented generation, so the AI reflects the course rather than generic chatbot output. After each activity, instructors receive per-question themes and misconceptions, a draft class-summary email, and a CSV export of all work. The platform will be used in a hybrid course this fall, and the showcase will demonstrate the live collaborative session, the AI feedback loop, and the instructor analytics surface.
Brainstorming with an LLM Student Activity
Kristin Stephens-Martinez, Associate Professor of the Practice, Computer Science
This presentation showcases a classroom activity that guides students in using large language models (LLMs) as a brainstorming tool for data science projects. Students are given a template LLM prompt, required to respond and critique the LLM’s ideas for multiple rounds, and asked to reflect on the experience. This activity encourages critical thinking, prompting skills, and responsible AI use. Attendees will gain a practical, ready-to-use activity that demonstrates how LLMs can support creative ideation while maintaining agency and accountability.
Creative Claude Code Projects by Non-STEM Undergraduates
Daniel Egger, Executive in Residence in the Engineering Graduate and Professional Programs
In Spring 2026, primarily non-STEM undergraduates in FECON 390 were introduced to the basics of Claude Code for the first time, then given free rein to develop their own art works and software applications. This short (5 minute) demonstration will highlight the remarkable range of creative output current Duke students are capable of with current LLM and multi-modal AI technologies as of May 2026.
Curating with AI: Virtual Exhibitions, Student Agency, and Critical Pedagogy in Art History
Nubia Nurain Khan, Graduate Student, Computational Media, Arts & Cultures
This demonstration presents a teaching experiment that integrated AI agents, virtual reality, and curatorial practice into an undergraduate global art history course, centered on Epochal Assemblages: Digital Curation, an interactive VR environment in FrameVR with embedded AI curatorial agents, designed to help students understand museum exhibitions as constructed interpretive arguments rather than fixed narratives. Developed for the 2026 CMAC PhD student exhibition and later adapted as a pedagogical model for ARTHIST 102D: Introduction to World Art History from 1200 to the Present, the project scaffolded a final assignment in which students created their own three-to-four-work virtual exhibitions in FrameVR. In their projects, students used AI in several ways, including as an organizational aid, as embedded knowledge agents, and, in some cases, as a tool for generating preliminary 3D models of artworks, podiums, and other digital curatorial elements. Although these AI-generated outputs were rough, they reveal how students use generative tools to solve concrete spatial, visual, and interpretive problems when given creative latitude. Together, the VR environment, the course scaffold, and these emergent student practices offer a humanities-centered model for teaching critical AI literacy through curatorial design.
Starship and Monastery: A Dual-Modal Framework for AI in Education
Ying Xiong, Senior Director, DKU Learning Innovation & Assessment, and Haiyan Zhou, Director of the DKU Center for Teaching and Learning
This showcase presentation introduces the DKU Dual-Mode AI Fluency Framework, a collaborative initiative developed by the DKU AI in Education Task Force. As GenAI reshapes higher education, the framework defines AI fluency as evaluative agency—the ability to make intentional judgments about when to engage with AI (“AI-on”) and when to prioritize human-driven thinking and learning (“AI-off”). The presentation first situates the framework within a broader institutional ecosystem that integrates AI fluency development, curricular design, policies and guidelines, and technological infrastructure to support a coherent and value-driven approach to AI in education. It then examines implementation through teaching practices and faculty development supported by DKU CTL, illustrating how AI can be strategically integrated or restricted to balance augmentation with intellectual independence. The session also invites discussion on how higher education institutions can meaningfully cultivate and assess AI fluency and evaluative agency.
Creating Engaging Interactive Demos with Marimo
Alex Steiger, Assistant Research Professor, Department of Computer Science
Marimo is a new Python notebook environment, similar in spirit to Jupyter, with features that make it especially useful for creating readable, interactive demonstrations for any course. In this showcase, we will highlight how an AI agent can be given direct access to a Marimo notebook and used to autonomously develop, revise, and debug instructional materials. We will present work-in-progress “skills” that guide the agent to consume existing course materials and convert them into Marimo notebooks with clear exposition, mathematical notation, and interactive pedagogical widgets. This workflow requires little to no programming experience.
Artificial Intuition? Why Contemporary AI Simulates, but Does Not Possess, Intuition
Rose Ansari, Graduate Student, Computational Media, Arts & Cultures
This presentation examines whether contemporary AI systems genuinely possess intuition or merely simulate intuitive behavior through statistical pattern prediction. Drawing from psychology, neuroscience, philosophy, and Bill Seaman’s Neosentience framework, the project evaluates AI systems using classical intuition-based psychological paradigms involving perception, moral judgment, social inference, and metacognition. The research argues that while AI can generate plausible intuitive responses, it lacks embodiment, affective physiology, lived experience, and genuine stakes in the world. The presentation also introduces Neosentience as a speculative educational and research framework that rethinks future machinic cognition through multimodal sensing, adaptive feedback, embodiment, and self-organizing systems.
The Evolution and Impact of Duke AI Health’s Scientific Writing for Staff Workshop Series
Whitney Welsh, Research Scientist, Duke AI Health
The Foundations of Scientific Writing for Staff workshop series is designed to provide a basic introduction to scholarly writing in the biomedical research settings, with a particular focus on equipping staff members with the knowledge, mentoring, and tools needed to participate actively in the entire process of developing, authoring, and publishing scholarly works such as presentation posters, abstracts, and peer-reviewed manuscripts. Over the course of the series, each participant developed a poster using formal scientific writing conventions, including topic selection and refinement, organization, content development, visual elements (including an introduction to basic elements of visual rhetoric and data visualization and display), ethical practices in co-authorship, use and management of references, acknowledgements, and other best practices. The workshops incorporated discussions of how (and how not) to use AI in scientific writing, though the emphasis was on learning by doing yourself the first time through in order to develop skills.
This poster will showcase the results of course evaluations from each of the three iterations of the workshop series, held between 2023 and 2025, to demonstrate that participants’ understanding of key concept related to scientific writing improved as a result of the workshop, and that the workshop meets a professional development need and that the content is relevant to participants’ work. The poster will also describe how participant feedback and instructor observations shaped the evolution of the workshop design to better meet the needs of staff, as well as lessons learned.
Helping Students Navigate AI Use Within Course Expectations
Yesenia Velasco, Lecturer, Computer Science
This poster presents a classroom approach for helping students reason through course-appropriate AI use in an introductory computer science course. Some students may be unsure how to apply course-specific AI policies to their own coursework, especially when those policies depend on context and differ across courses. Rather than presenting AI guidelines as a checklist, the approach uses concrete examples to discuss when AI may support learning and when it may conflict with course goals or collaboration guidelines. Examples include using AI to generate practice problems, clarify concepts, or think through problem solving, while distinguishing those uses from ones that replace the student’s own reasoning or work. By making expectations more visible and concrete, this approach aims to give students a clearer basis for deciding how to use AI in coursework, and attendees will see examples they can adapt when communicating AI expectations to their own students.
AI, Textbook, and Peer-Reviewed Journal Comparison with Bias Reflection: Enhancing Critical Appraisal and Evidence-Based Practice in Emerging Technologies
Khaled W. Bader, Assistant Clinical Professor, School of Nursing
The increasing use of artificial intelligence (AI) in nursing education provides valuable support for learning, content generation, and academic writing, but it also raises concerns about inaccuracies, incomplete information, and bias that may affect clinical reasoning and evidence-based practice. This quality improvement project aims to develop and implement an AI/textbook comparison and bias reflection assignment to strengthen nursing students’ critical thinking and ability to evaluate AI-generated clinical information. Using the PARIHS framework, a quasi-experimental pre–post design, and the PDSA cycle, the project will assess the effectiveness of the assignment through student feedback, reflections, and performance outcomes. Nursing students enrolled in the CM3 course at Duke University School of Nursing will compare AI-generated disease-specific content with textbooks and peer-reviewed sources to identify discrepancies, evaluate reliability, and create evidence-based care plans. This intervention promotes ethical and responsible AI use while preparing future nurses to integrate AI safely into clinical practice and patient care.AI, Textbook, and Peer-Reviewed Journal Comparison with Bias Reflection: Enhancing Critical Appraisal and Evidence-Based Practice in Emerging Technologies
ICYMI: AI, Intellectual Property & Marginalized Groups – A Panel Discussion in Remembrance of Judge James Donald Smith, L’86
Geovanny E. Martinez, Senior Lecturing Fellow and Executive Director for the Center on Law, Race & Policy
In March, the Center on Law, Race & Policy at Duke Law hosted “AI, Intellectual Property & Marginalized Groups – A Panel Discussion in Remembrance of Judge James Donald Smith, L’86.” You can watch the recording here: https://youtu.be/CeZ9WmmDi4c. This event was in honor of the late Judge James Donald Smith, a Duke Law alum, who served as the Chief Administrative Patent Judge at the United States Patent and Trademark Office and was responsible for establishing the Patent Trial and Appeal Board, and he also held prominent positions in academia and led in-house intellectual property teams for several multinational corporations during his illustrious career. The panel discussion focused on the potential effects of AI on artists’ rights, intellectual property law, and marginalized groups. The event featured: Duke Alum Brent Clinkscale (Duke A.B.’83 and L’86)– (Moderator) Independent Arbitrator, Mediator and Litigation Consultant, Clinkscale Global ADR; Duke Alum Detavio Samuels (Duke B.A’02) – CEO, Offscript Worldwide and REVOLT; Judge (Ret.) Scott Boalick – Former Chief Administrative Patent Judge, USPTO; Joseph Drayton – Litigation Partner, Proskauer Rose LLP; Olufunmilayo Arewa – Professor of Law, George Mason University Antonin Scalia School of Law; and Duke Law Professor Arti Rai – (Offering Introductory Remarks) Elvin R. Latty Distinguished Professor of Law and Co-Director of the Center for Innovation Policy at Duke Law.