8:30am - 9:30am

9:30am

9:45am - 10:30am

10:40am - 11:30am

11:30am - 12:20pm

12:30pm - 1:55pm

3:00pm - 3:45pm

Andrew Abernathy, Rosemary Avance

Using a Collaborative Online International Learning (COIL) framework, the authors implemented a cross-cultural experiential learning project in which 73 journalism and mass communication students from the U.S. and Mexico co-created a generative AI social contract for use in their courses. This approach enabled students to engage in cross-cultural collaboration, structured deliberation, and democratic voting to define the parameters of acceptable and prohibited uses of AI. Outcomes suggest that this intervention promotes student buy-in; increases awareness of AI uses, risks, and opportunities; and provides a replicable framework for AI policymaking to promote transparency, decrease procedural anxiety, and support human learning. In addition to reporting results, our presentation will outline implementation practices and provide materials that other faculty can adopt and/or modify.

Lu Zhai

Wildfires in the United States are increasing in intensity and frequency, generating rapidly expanding and complex datasets that remain difficult to analyze and communicate efficiently. This project addresses this gap by leveraging advances in artificial intelligence (AI), particularly large language models (LLMs), to enable fast, interpretable, and data-driven insights from wildfire data. We propose to develop FireGPT, a multimodal, retrieval-augmented AI platform that integrates curated Earth observation datasets of wildfire with quantitative tools for analysis, visualization, and transparent reasoning. By transforming complex data into accessible knowledge and actionable insights, this project will advance wildfire science, support evidence-based decision-making, and foster inclusive, cross-sector collaborations that strengthen resilience to wildfire risks.

Bobby Reed

The Gang of Four no longer resonates with Comp Sci majors the way it once did. That doesn't mean that programming design patterns aren't more important than ever. In the Spring of '26, Oklahoma City University morphed their Game Programming class to teach classic behavior and creation design patterns through the context of game design in a Unity/Monobehavoiur/C# tech stack. In this lightning talk, the outcomes and learns of this course will be shared. Particpants will be invited to help shape the next round of this course as well by providing feedback via a post-talk survey on how design patterns have affected their professional career.

Jenny Bodenhamer

In late 2024, Oklahoma State University Library's Digital Resources and Discovery Systems (DRDS) department utilized the OpenAl API (GPT-4 Turbo) to generate 20,000 abstracts for a corpus of historical, digitized theses and dissertations. Recognizing the ethical risks of deploying unmediated Al-generated metadata, DRDS conducted a systematic review of 400 sample abstracts. This audit revealed that the generative process occasionally replicated harmful legacy language and pejorative terminology present in the original historical texts.

To address these findings, DRDS implemented a secondary evaluative workflow using the OpenRefine Al Extension. Utilizing the natural language processing of Al, this workflow performs content analysis on all the abstracts, flagging instances of exclusionary language for manual remediation. By integrating this computational ""human-in-the-loop"" oversight, the library establishes a framework for critical cataloging and reparative description.

Andrew Doust

Faculty Working Group on AI Use, College of Arts and Sciences, Oklahoma State University

Universities are rapidly deploying AI integration initiatives in response to student and employer demand, yet a critical gap remains: faculty pedagogical expertise has been largely absent from decisions about where AI enhances learning versus where it undermines essential cognitive development. Many faculty have concerns that AI will diminish student critical thinking skills, yet few faculty feel that their institutions have properly prepared them to address AI's pedagogical challenges and opportunities. A top-down approach risks implementing AI tools without preserving the human-centered competencies that are fundamental to student learning.

To address this gap, a semester-long faculty working group was convened, comprised of twenty instructors from thirteen departments spanning the disciplinary breadth of the College of Arts and Sciences. Through facilitated discussions and exercises, the group tackled two essential questions: (1) What specific cognitive skills and competencies should remain human-centered and unaided by AI? and (2) How can faculty effectively teach with and about AI while preserving essential competency development?

The working group produced three concrete deliverables. First, we defined critical thinking and its necessary requirement for cognitive effort and intellectual struggle (“cognitive friction”) as central to human-centered competencies. We identified specific aspects of critical thinking with varied implications in each discipline, such as working with evidence, creating and evaluating arguments, critical reading, communication, and synthesis and application. Second, we compiled examples of course implementation of AI-related assignments in courses ranging from those that expressly restrict student AI use to prioritize human-centered learning to those in which AI use is central to student learning. Third, we provided recommendations to create a system of discipline-specific faculty fellows to provide guidance and support to faculty as they incorporate or restrict AI tools in their pedagogical approach to assignments and courses.

This faculty-driven approach offers institutions an alternative to efficiency-focused AI adoption: a pedagogically grounded framework that positions faculty expertise at the center of AI integration decisions and recognizes that appropriate AI use varies across disciplines. Our work demonstrates that when faculty lead with pedagogical principles, AI becomes a tool for enhancing, rather than replacing, the essential human competencies that define deep learning in higher education.

Heather Stewart

More and more, faculty across academia are being encouraged to integrate AI tools into their pedagogy (whether or not such AI integration would benefit their course pedagogically or not). Such AI integration across disciplines, faculty are told, is essential, given the technological demands and realities of the workforce our students will be entering into post-graduation. While it is of course true that some familiarity with and literacy around AI will be essential for many students entering many careers post-graduation, this presentation argues that there are nevertheless strong moral and pedagogical reasons to retain the option for students to opt out of the use of AI tools in their classes (e.g., to not require students to engage with AI tools when they have strong reasons against doing so).

First, students might value other things about their time in university beyond job preparation and training (e.g., even if it is true that practice with AI tools is beneficial for their future career readiness, they might value other things more, e.g., practicing certain skills in a way that is not aided (or impeded) by AI). Second, it isn’t always clear that the ways faculty would be integrating AI into their classes is in fact beneficial with respect to what future employers may actually desire  (e.g., most faculty aren’t technical experts in AI and aren’t well equipped at this stage to provide meaningful AI training to students). They are, however, content experts in their respective disciplines and are likely well positioned to teach students the high-level, abstract, and big-picture kinds of thinking that employers might actually want, especially as AI tools become increasingly prevalent and more sophisticated). Third, some students might desire an AI-free education and/or to insulate their cognitive, intellectual, creative, emotional, and interpersonal skills from AI (as much as possible) which is worth respecting and creating pathways for them to achieve. Fourth, some students might have serious moral, environmental, or political objections to using some or all AI tools (e.g., they might want to boycott certain AI companies that have particularly damaging impacts on politics, the environment, etc., or who contract with the US Department of Defense. Or, they might have concerns regarding data privacy that mean they don’t want to engage with such tools). Such serious moral and political commitments should be respected. Finally, at this stage students will not have equal access to AI tools (e.g., some students will have access to paid AI tools while others will not).

For all of these reasons, I argue that it is imperative that students retain the right to opt out of AI use in our classes (unless it is absolutely essential to the content of a particular course, e.g., in Computer Engineering or cognate fields).

Kathryn Weinland, Jennifer Glenn

The increasing availability of artificial intelligence (AI) tools in educational settings raises important questions about their impact on student engagement and learning. This study examined whether low-stakes gamification, implemented through the platform Kahoot!, may serve as a buffer against in-class AI use. Data were collected in Fall 2025 from undergraduate engineering and psychology courses. In some course sections, instructors incorporated Kahoot-based review questions, while others did not. Students self-reported whether they used AI tools to assist with answering in-class questions.

This study compares (a) AI use in classes that used Kahoot versus those that did not, (b) differences in AI use across disciplinary contexts (engineering vs. psychology), and (c) accuracy of responses between students who reported using AI and those who did not. It is hypothesized that gamified, low-stakes environments may reduce reliance on AI by promoting real-time engagement, peer collaboration, and critical thinking. Findings will contribute to understanding how instructional design strategies can shape student behavior and ethical AI use in the classroom.

Dani Kirsch, Kathy Essmiller

The utility of generative artificial intelligence (genAI) tools is in part dependent upon their training datasets. The contents of these training datasets are one of many points of contention with genAI, illustrated by allegations of piracy and copyright infringement, accusations of bias, and attempts to control the ingestion of digital content into genAI databases. Individual researchers and educators are scattered along the continuum from complete opposition to full embrace of genAI and the content it produces. These perspectives encompass not only personal decisions and opinions on the use (or non-use) of tools but also the willingness to “feed” these tools with scholarly outputs such as articles, books, and data.

This presentation aims to overview some of these major points of tension from the perspective of open scholarship. We will briefly discuss copyright and licensing to foreground the conversation around author rights, reuse permissions, and reciprocity. We will then present perspectives on open scholarship, focusing particularly on motivations for publishing content openly and how these motivations can clash with expectations and behavior from genAI tools and companies. Finally, we will share a few approaches that have been proposed to tackle specific problems in these areas.

Shuai Zhang, Jaina Agan, Mohammad Aazir

There have been many different tools used to identify individual animals; all with their own advantages/disadvantages. The disadvantage that is present in nearly all previous marking methods is that it is stressful for the animal as it requires capture and handling to mark them. Modern technology through deep learning frameworks allows us the opportunity to use publicly available data on specific species to train models that specialize in identifying individuals. In this study, we are attempting to train a model to identify specific individuals in the species Crotaphytus collaris. We collected photos of Crotaphytus spp. from iNaturalist for 100+ observations to represent the genus’ morphological diversity. This diversity introduced significant challenges that were addressed by implementing a pixel-wise image segmentation approach using YOLO11. Over 500 images were annotated with body contour polygons to train, validate, and test our deep learning model for segmentation of Crotaphytus spp. from the background. Results demonstrate that, even in cluttered natural environments with occlusion, our model achieved an average precision (AP50) of 91.3%. Our next steps are to refine the model to work with a larger population of lizards and start training a model for individual identification.

Frederic Murray

This presentation addresses the emerging business dilemma posed by artificial intelligence (AI) in higher education, with a particular focus on the evolving role of academic libraries. The central problem lies in the growing tension between the commercialization of AI technologies and the core academic values of equitable access, intellectual freedom, user privacy, and the integrity of the scholarly record. As higher education institutions integrate AI-driven systems into research, teaching, and administrative functions, librarians are increasingly called upon to navigate these competing priorities.

For the purposes of this discussion, “artificial intelligence” refers to generative and predictive systems capable of producing text, synthesizing information, and supporting decision-making processes. The “business dilemma” describes the structural conflict between proprietary AI platforms—developed and controlled by commercial vendors—and the academic mission to ensure open, transparent, and sustainable access to information resources.

Contemporary examples illustrate these tensions. Subscription-based AI research tools risk exacerbating inequities among institutions, while the opacity of training data introduces concerns related to bias, accuracy, and accountability. Instructional applications, including AI-assisted writing tools, further destabilize conventional understandings of authorship and academic integrity.

The implications for academic libraries are substantial. Librarians must adapt collection strategies, engage in more complex vendor negotiations, and expand information literacy instruction to encompass AI fluency and critical evaluation of machine-generated content. More broadly, higher education must determine whether its adoption of AI will reinforce commercial dependency or support the development of ethical, open alternatives. This presentation argues that librarians are essential stakeholders in shaping institutional responses, ensuring that innovation remains aligned with enduring academic values.

Matt Vassar, Dursun Delen, Joshua Habiger

The deployment of artificial intelligence in healthcare is accelerating faster than the systems we rely on to evaluate it. In Oklahoma, the Rural Health Transformation Program is rolling out AI-enabled clinical documentation tools to providers across the state. This is an important and promising investment. Separately, Oklahoma is one of six CMS pilot states where AI is currently making Medicare prior authorization decisions under the WISer Model, a live deployment that Oklahoma’s own medical community has publicly questioned. At the same time, we lack a consistent way to determine whether these tools are producing reliable and trustworthy outputs. The gap is not simply about ethics in a general sense. It is a concrete and measurable technical issue. AI systems trained on published biomedical literature inherit the same selection biases that shape the scientific record itself. When models learn from that record, they absorb its blind spots. Null findings, adverse events, and research involving rural or underserved populations are often underrepresented. The model does not know what was never published, and that absence quietly shapes its conclusions. This becomes especially important in rural settings. A provider in Oklahoma may receive an AI-generated clinical recommendation without immediate access to specialist support. In that moment, there is no built-in way to assess whether the recommendation reflects strong underlying evidence or a polished synthesis of incomplete data. We are developing a coordinated response across OSU to address this problem. Vassar will outline the evidence integrity challenge in the Oklahoma context, drawing on a large body of work examining adverse event underreporting. He will also describe a summer initiative in which 30 OSU medical students are annotating hundreds of clinical trial records to create a validated training dataset. Delen will introduce a federated agent architecture that allows local summarization and classification while keeping identifiable clinical data in place. This approach fits the privacy and infrastructure realities of rural healthcare. Habiger will present a statistical framework that formalizes how AI systems may amplify publication bias, along with a conformal prediction approach that provides calibrated uncertainty rather than overconfident outputs. We close by outlining how AI adoption and AI validation can function as complementary investments at OSU. We also extend an invitation to collaborate across disciplines, recognizing that this challenge sits at the intersection of data science, clinical care, and public trust.

Madison Hastings

As artificial intelligence (AI) tools such as ChatGPT become increasingly integrated into research and educational workflows, health sciences librarians are essential in guiding their ethical and effective use. This presentation explores how AI can support the development of rigorous, reproducible search strategies while maintaining standards of transparency and information literacy.

The session will demonstrate how structured prompting techniques can be used to translate clinical and research questions into database-specific search strategies across PubMed, Web of Science, and Embase. Attendees will learn how to use AI to generate keywords, controlled vocabulary (including MeSH and Emtree), and Boolean logic, as well as how to iteratively refine outputs for precision and recall. Emphasis will be placed on evaluating AI-generated content for bias, gaps in retrieval, and reproducibility concerns.

In alignment with the conference theme, this presentation also addresses practical challenges, including ethical considerations, data privacy, and the risks of overreliance on AI tools. Strategies for integrating AI literacy into library instruction and developing institutional guidelines for responsible use will be discussed.

By the end of this session, participants will be able to (1) apply structured prompting techniques to generate database-specific search strategies, (2) evaluate AI-generated search outputs for accuracy, bias, and completeness, and (3) identify key ethical and procedural considerations when using AI in health sciences research workflows.

Brandon Halte

Douglas Hickey

Generative artificial intelligence is frequently championed for its capacity to synthesize vast amounts of information and provide immediate assistance. However, prioritizing computational efficiency over the phenomenological process of learning fundamentally misunderstands how human learning optimally develops. In the realm of AI facilitated learning, empirically informed cognitive science demonstrates that durable education and knowledge retention requires frequent dynamic adjustment to dial in a desirable difficulty. This poster addresses a profound vulnerability in current AI interactions. Large language models are commercially fine-tuned for algorithmic sycophancy and user compliance and engagement. This relentless sycophancy and engagement farming strips away the essential cognitive friction required for rigorous logical and moral reasoning. When an AI instantly resolves complex academic queries, it bypasses the psychophysics of human intention and learning. Specifically, it prematurely relieves the cognitive load required to translate raw streams of cognition into a logically mapped argument. By removing the necessary struggle of articulation, the model breeds intellectual overconfidence rather than genuine understanding.

To harness AI responsibly, I propose that cognitive friction is a feature, not a bug. Rather than treating AI as an automated oracle, educators must construct environments where AI functions as an adjustable cognitive scaffold. This project introduces a programmed epistemic resistance framework that is implemented via a dialectical scaffolding agent. Designed to deliberately slow students down, this agent refuses to generate prose. It instead forces users to build argument maps and empirically justify premises prior to drafting. This programmed resistance also serves as a universal accessibility tool. In an attention economy that constantly fractures focus, the agent provides momentum kindling. It offloads the paralyzing burden of task initiation for students while strictly and strategically preserving the cognitive load required for actual academic synthesis.

Arati Poudel, Dr. Phil Alderman, Dr. Brett Carver, Dr. Meriem Aoun, Grishma Ojha

Barley Yellow Dwarf (BYD) is one of the major viral diseases of wheat which is caused by multiple viruses from the Luteoviridae family and transmitted by aphids. BYD resistance genes identification is difficult, because the nature of the resistance is quantitative and polygenic. However, high-throughput phenotyping can help. Thus, this research aims to develop a set of UAS-based features that best predict BYD severity and help breeders for large scale selection of BYD resistance in wheat. For data collection, the multispectral UAVs were flown periodically during the years 2023 and 2024 on wheat breeding trials at the OSU Agronomy Research Station, Stillwater. These UAV images were used to calculate Digital Elevation Models (DEMs) and six vegetation indices (VIs); three using RGB sensors only, and the other three using multispectral sensors. Four levels of statistics — mean, standard deviation, skewness, and kurtosis, were calculated for each plot to capture the subtle variations in reflectance that might signal disease. DEMs were used to estimate height thresholds (full height without thresholding, 90% plant height, and 75% plant height) for filtering VI pixels to see if disease patterns are more visible at certain canopy layers. Different combinations of sensors, statistics, and height thresholds were used as factors for analysis to fit Bayesian Ordinal Logistic Regression model for predicting BYD severity. RGB data using all four statistical descriptors and no height thresholding, gave the best predictions. Adding statistical descriptors consistently improved macro-recall and macro-precision for both MS and RGB data. Height-based filtering may not be necessary for BYD detection, meaning that disease symptoms seem to appear across the whole canopy. Pixel-level statistics contributed more to disease prediction than height-based filtering. So, RGB data with the extensive use of statistical descriptors were the top performing combinations among different combinations used.

Yomal Wijesiriwardena, Christopher J. Fennell

The incorporation of self-learning and high throughput algorithm is introducing a conceptual shift in the conventional scientific approach within computational chemistry. This study introduces a novel automation pipeline combining neural network models and HPCs for the systematic screening of huge numbers of molecular parameters. Machine learning (ML) can be quite effective computationally, yet it poses several challenges, one of which is referred to as "black box" because it usually results in force fields which might be mathematically converged but physically irrelevant. To address the above issues, the proposed methodology involves the use of physics-related constraints in addition to machine learning to develop more accurate force fields. The introduction of phase classification will act as an excellent filter and make sure all models used in simulations are at least liquid phase stable. On the other hand, an uncertainty driven active learning algorithm will allow us to screen for the best models in terms of energy and speed in an intelligent way. This methodology is designed to impose tough physical constraints by employing Weighted Absolute Percentage Error as the objective function. The tough validation we carry out gives us irrefutable proof that our AI-based model can describe the complex tetrahedral hydrogen bonding in water.

Kristyn Lacki

Design information disclosure in additive manufacturing process data has become an important consideration, as machine learning models may be used to reverse engineer the design information. In this research, thermal image data collected from a laser-based powder-bed fusion manufacturing process will be used to train a machine learning model to predict instantaneous printing orientation. Using this model, test values may be compared to collected data to observe similarities and, by extension, the propensity of the model to extract design information. Furthermore, certain conditions may indicate the potential for the model to prove successful in reconstructing the print path throughout the design process. The proposed framework will result in a better understanding of the extent to which the design information can be predicted using thermal images, leading to a quantifiable evaluation of the vulnerabilities in the security of product design data.

Anastasiia Oskolkova

Juvenile-onset recurrent respiratory papillomatosis (JORRP) is a chronic airway disease characterized by recurrent papilloma growth that often requires repeated surgical removal. Inhalation therapy may help reduce papilloma recurrence, but effective treatment depends on precise drug delivery to specific airway sites while minimizing off-target deposition. Because regional drug deposition cannot be readily measured in vivo, computational fluid-particle dynamics (CFPD) is a valuable tool for predicting site-specific aerosol delivery. However, besides forward prediction, realizing personalized inhalation therapy also requires to identify the key parameters of inhalation therapy that can achieve prescribed deposition targets. Exhaustive or iterative CFPD-based solution search over parameter combinations is computationally expensive and often impractical for real-world treatment planning, motivating the development of a data science-enabled smart inverse design methodology.

This study developed an integrated computational framework for personalized inhalation treatment design in JORRP by teaming CFPD simulation and machine learning (ML). CFPD-generated simulation data were used to link inhalation system parameters (release position, inhalation flow rate, particle diameter, and orifice aspect ratio) to regional deposition outcomes in the glottis and larynx. ML models were also trained to inversely predict inhalation parameter settings from prescribed target deposition profiles, while a forward surrogate model estimated the resulting deposition and supported candidate evaluation within a closed-loop inverse-forward prediction pipeline.

The proposed methodology aims to address the non-uniqueness of the inverse mapping: multiple inhalation parameter combinations can produce similar deposition outcomes. To manage this challenge, the framework evaluates inverse predictions in deposition space rather than only in parameter space, enabling model selection based on how well predicted designs reproduce intended deposition targets. The framework also supports mixed parameter types through a sequential prediction strategy.

On a balanced CFPD-simulated dataset, the proposed inverse design approach achieved surrogate-assisted reconstruction of prescribed regional deposition targets, with normalized mean square error below 0.1 and R^2 above 95%. Such results also indicate that the proposed methodology is capable of reproducing target deposition patterns within the learned surrogate modeling environment. Thus, the proposed methodology has great potential to provide an effective and efficient solution for simulation data-informed inverse design in personalized treatment planning and related broad engineering applications.

MenyeneAbasi Obong

The rapid integration of Generative Artificial Intelligence (GenAI) into higher education is transforming not only students’ writing but also instructors’ roles and identities. This study examines how five (5) Graduate Teaching Assistants (GTAs) in a college writing program at a North Central University, in the United States, negotiate their professional identities in response to students’ varied uses of GenAI tools (in writing classes) and writing program AI policies. Grounded in an interpretive framework and informed by Darvin and Norton’s (2015) model of investment, the study draws on semi-structured interviews and AI policy documents from writing programs to explore how identity, ideology, and power intersect in writing classrooms where students are unevenly mediated by GenAI tools. Findings reveal that GenAI tools function as more than a technological tool; they serve as mediational tools that reshape classroom authority, authorship, and pedagogical decision-making. First, GenAI tools intensify ethical tensions around authorship and academic integrity, positioning instructors as gatekeepers of legitimate participation in writing tasks. Second, AI policies from some writing programs within the university under investigation redistribute power by formalizing teacher authority in writing classes, and at the same time, increasing surveillance and accountability. Third, instructors oscillate between roles as mediators, evaluators, and advocates, as they make sense of the competing expectations from students and writing programs. This study contributes to ongoing interdisciplinary conversations about AI in education. It also extends Darvin and Norton’s model of investment by demonstrating that teachers, like learners, negotiate identity, particularly as AI reshapes authority, authorship, and pedagogical practice in writing classrooms. Implications include the need for policy (re)design and teacher education programs that account for instructors’ evolving roles in learning environments, in the GenAI era.

Amelia Sanders

My presentation focuses on the use of Khanmigo in the classroom. This is an Artificial Intelligence program from Khan Academy that provides a multitude of resources for teachers to use as they plan lessons, write IEPs, build quizzes and create exit tickets. I recently used the Fun Class Summary Poem with my Literacy Tutoring student during our sessions and they loved it! My presentation would include pictures of the poems we created using the tool, as well as how I used the poem generator to build on the skills we practiced in our sessions.

Harlie White, Brennan Cowan, Jaina Agan

Eastern collared lizards (Crotaphytus collaris) overwintering activity have not been well studied or observed, rather most studies do not consider the implications of overwintering in their studies nor how climate change may be impacting the activity. The purpose of this study is to study when lizards are coming and going from burrows throughout the year and to learn more about the overwintering activity of Eastern collared lizards, distinctly how often they move and where they move to and from. To monitor activity, we installed a RFID reader module and camera to multiple Raspberry Pi computers and programmed the Pi to take a picture every time the RFID reader reads a tag. We then placed the readers in weather resistant containers in front of burrows and caught lizards at the field site. We toe-clipped and implanted each lizard with an RFID tag subcutaneously. The data collected will be utilized to determine collared lizard’s activity seasonally and to set a baseline for activity in response to future climate change.

Sharad Puri

In this work, ZnO nanorods were grown on vertically aligned and randomly aligned silica nanosprings using the hydrothermal method. The initial step was the deposition of a ZnO seed layer by atomic layer deposition to promote nucleation. For hydrothermal growth, equimolar (0.2 M) solutions of Zinc nitrate hexahydrate and hexamethylene tetraamine prepared in DI water were used. The ZnO NR grown on the VANS were flower-like clusters, while for the RANS, the ZnO NR grew radially outward from the individual nanosprings. The lengths and diameters of ZnO NR grown on VANS and RANS were 175 and 650 nm, and 35 and 250 nm, respectively. Scanning electron microscopy confirmed the formation of ZnO nanorods, while X-ray diffraction and Raman spectroscopy verified that they have a hexagonal wurtzite crystal structure with preferential growth along the c-axis. X-ray photoelectron spectroscopy, in conjunction with in vacuo annealing, was used to examine the surface electronic structure of ZnO nanorods and defect healing. Photoluminescence of the ZnO nanorods indicates high crystal quality, as inferred from the weak defect band relative to strong excitonic band edge emission.

Abhilash Minukuri, Sai Tharuni Samineni

Artificial intelligence is becoming a common source of help for students learning programming, but fast AI-generated answers do not always support deep understanding. In introductory programming, students often need timely feedback on syntax, logic, and debugging errors, yet unstructured AI support can encourage overreliance and reduce productive problem solving. This project explores a human-in-the-loop approach to AI-supported personalized formative feedback for novice programming learners. Rather than giving complete solutions immediately, the proposed approach emphasizes staged feedback that helps students identify errors, reflect on their thinking, revise their code, and build greater independence over time. The framework focuses on six main elements: learner-state traces, feedback target, feedback progression, gated scaffolding, human oversight, and risk-tiered safeguards. The goal is to support learning in ways that preserve student agency while still providing timely and personalized guidance. This work contributes a conceptual foundation for designing AI-supported feedback systems in introductory programming and lays the groundwork for a future prototype and empirical study.

Shafi Al Salman Romeo

Atmospheric entry is one of the most critical phases of any space mission. As a capsule decelerates from hypersonic speeds, it oscillates due to the complex interactions between the capsule and the flowfield around it, and if those oscillations grow beyond a critical threshold, parachute deployment fails and the mission is lost. For crewed missions like NASA's upcoming Artemis II, understanding and predicting this dynamic instability is not just an engineering challenge. It is a matter of human survival. Traditional methods for characterizing entry capsule stability rely on wind tunnel testing, ballistic range experiments, and high-fidelity computational fluid dynamics (CFD) simulations, all of which are expensive, time-consuming, and difficult to generalize. More critically, existing data reduction techniques struggle to quantify the uncertainty in their predictions, leaving mission designers with incomplete risk assessments. This work presents a modular, AI-driven Nonlinear Parameter Estimation (NPE) framework that extracts dynamic stability coefficients directly from CFD simulation data and, crucially, tells you how confident to be in those predictions. The framework combines a physics-informed neural network with a Markov Chain Monte Carlo (MCMC) algorithm. The neural network is trained on a representative physics equation rather than expensive CFD data, and it intelligently initializes the MCMC sampler, dramatically accelerating convergence and improving robustness. The MCMC method then estimates both the pitch damping coefficients and their associated uncertainties from free-flight CFD trajectory data. Applied to a scaled ballistic range model of the Genesis capsule at Mach 1.44, the framework successfully recovered dynamic and static stability coefficients with reconstructed trajectories in excellent agreement with high-fidelity CFD truth data. The approach is generalizable, computationally efficient, and directly applicable to mission-critical aerodynamic database generation, offering a new standard for data-driven stability analysis in the era of crewed planetary exploration.