The annual MSOE Professional Development Fall Forum is Thursday, Nov. 12 from 12 to 1 p.m. in Diercks Hall Atrium. The forum showcases the exciting scholarly projects faculty are leading across campus. These scholarly projects are supported by Summer Development Grants. Learn more in the abstracts below.
Summer Development Grants
1. Development of a Structured Medication Calculation Remediation Program for Nursing Students
Drs. Stephanie Bruce and Sarah Arvelo (School of Nursing)
Medication dosage calculation is a foundational skill for safe nursing practice; however, nursing students enter programs with varying levels of mathematical preparation, confidence, and calculation proficiency. To address these challenges, a structured medication calculation remediation program was developed within Canvas to provide targeted support for students requiring additional practice and skill development. The project resulted in the creation of 12 self-paced remediation modules covering foundational mathematics, dosage calculations, dimensional analysis, intravenous medication calculations, and clinical application problems. In addition to calculation-focused content, the program incorporates resources addressing test anxiety and confidence-building strategies to support student success more holistically.
Each module utilizes a standardized structure that includes instructional content, worked examples, guided practice activities, mastery assessments, and reflective learning exercises. The program was designed using a scaffolded approach that guides students from foundational concepts to increasingly complex clinical calculations while promoting mastery learning and patient safety.
The remediation framework has been implemented within the MSOE School of Nursing and made available to nursing students as an independent learning resource. Institutional Review Board approval was obtained to evaluate program effectiveness and support future dissemination. This poster will describe the development process, implementation strategies, lessons learned during program creation and beta testing, and plans for formal evaluation of student outcomes during Fall 2026.
2. Direct Visualization of Fluid Inclusions Formation during Evaporation-Driven Crystallization
Dr. Xiaoyi Hu (Mechanical and Industrial Engineering)
Evaporation-driven crystallization of solute-containing microdroplets can produce complex internal structures, including fluid inclusions that may retain liquid after bulk evaporation. This project investigated microdroplet evaporation, crystallization, and potential fluid-inclusion formation using bottom-view microscopy. Preliminary data were acquired on droplet evaporation dynamics and NaCldroplet evaporation at varied initial salt concentrations. The experiments captured droplet shrinkage, crystallization, and resulting crystal morphology, with internal features identified as candidates for fluid inclusions. In parallel, preliminary experiments evaluated the survival of E. coli under extreme environmental conditions relevant to concentrated and evaporating droplets. These results provide a foundation for future studies of the transport processes governing fluid-inclusion formation and their potential role in microbial survival during droplet evaporation.
3. A Scaffolded Pedagogical Framework for Teaching Critical Evaluation and Direction of AI Coding Agents
Dr. James Lembke (Diercks School of Advanced Computing)
AI coding agents that autonomously read, modify, and commit code are reshaping software development, yet no established pedagogy exists for teaching students to direct these tools effectively. A scaffolded undergraduate course progresses from unstructured AI interaction through structured workflow execution to independent environment design. Students learn to provide explicit project context through grounding files, constrain agent behavior with behavioral gates, evaluate output using a three-question framework, and assess their own process rather than the AI's artifacts. The framework is discipline-agnostic: its core mechanisms (context specification, output evaluation, process portfolios) transfer to any setting where students interact with AI tools.
4. Development of a novel CNT-based plasma–UV source for undergraduate education and applied research
Dr. Nazieh Masoud (Physics and Chemistry)
This project involved the design, construction, and testing of a novel, reconfigurable plasma–UV source using carbon nanotube (CNT) as the electron emitters and ceramic thin films as a barrier between the CNT and the ambient air. The source was built and experimentally tested to verify its operation and its ability to generate plasma and UV emission. Its modular design allows the system to be adapted for different experimental and research applications and provides opportunities to incorporate diagnostics such as optical emission spectroscopy and electrical characterization. The completed source will be implemented in the Biophysics course, providing students with hands-on experience using a research-level experimental system. Students will investigate plasma and UV interactions with biological materials while developing experimental, data-analysis, critical-thinking, and scientific communication skills.
The source also establishes a new platform for Senior Design projects, independent studies, and undergraduate research courses. Its flexibility creates opportunities for interdisciplinary projects involving applications such as UV-based sterilization, microbial inactivation, and plasma–surface interactions with biomaterials. Overall, the project expands undergraduate research capabilities and creates a new connection between classroom learning, hands-on experimentation, and interdisciplinary research.
5. Robust Synthetic Data Generation
Drs. RJ Nowling and Jeremy Kedziora (Diercks School of Advanced Computing)
Large language models (LLMs) offer a powerful approach to synthetic data generation, but naively generated individual outputs do not necessarily overlap with the support of the distribution of real data. We propose a probabilistic filtering method that uses a classifier trained to distinguish real from synthetic documents to guide acceptance of LLM-generated text. The method operates on complete documents and requires neither generator fine-tuning nor access to model internals, making it compatible with existing provider APIs. We formulate our approach as a minimax problem over a class of detectors, characterize the solution of this minimax problem, and show that distributional overlap can be realized through rejection sampling. We evaluate the approach across four LLMs and four text datasets using logistic regression with binary bag-of-words features. Our results show that unfiltered synthetic documents are readily distinguishable from real documents, with baseline ROC AUC scores ranging from 0.812 to 1.000. Probabilistic filtering generally reduces distinguishability, although improvements vary across models and datasets. We also investigate how classifier trainingset size and oversampling of limited real data affect probability estimates and filtering performance. These results support classifier-guided document selection as a practical approach to improving synthetic text fidelity.
6. AI-Augmented Motion Capture for Enhanced Biomechanical Analysis
Dr. Ahmed Sayed (Electrical, Computer and Biomedical Engineering)
Conventional biomechanical analysis depends on laboratory motion-capture and force-platform systems, which limits its use in field and outdoor settings. This project explored AI-based alternatives by building a kinematics-driven activity classification pipeline. Kinematic time series data were encoded as images and classified using the ConvNeXt network through transfer learning. Using public datasets that span joint-angle, optical marker, and IMU data, the model classified four activities: baseball pitching, baseball hitting, running, and walking. It achieved an overall 99.36% subject-independent accuracy across 1,400 trials with a leave-one-subject-out evaluation study. An originally planned marker-free validation step was not possible because the upgraded motion tracking lab system, we have on-campus, lacked true markerless tracking, which is the main limitation that I faced in this summer project. A video-based extension of the pipeline was also designed, relying on alternative tracking software, such as OpenCap and Theia3D. Together, this work establishes a foundation for future markerless motion analysis and course integration.
7. Hidden Risk Phenotypes in Noninvasive Respiratory Support: A Two-Dataset Study
Dr. Yiyang (Ian) Wang (Diercks School of Advanced Computing)
Patients receiving noninvasive respiratory support follow diverse clinical trajectories, and cohort-level analyses may obscure subgroups with different risks of treatment failure. This student-faculty summer project develops a reproducible pipeline for discovering patient phenotypes and evaluating risk models using MIMIC-IV and AmsterdamUMCdb. We defined cohorts around the initiation of respiratory support, constructed baseline and post-initiation physiological features, and applied unsupervised clustering and supervised classification. In AmsterdamUMCdb, two baseline phenotypes characterized by hypoxemic and hypercapnic presentations had similar 72-hour intubation rates (17.5% and 16.7%). A phenotype based on six-hour physiology showed greater separation in subsequent intubation rates (18.1% versus 12.0%). In MIMIC-IV, a supervised model achieved an AUROC of 0.741 at baseline and 0.757 at six hours. A timing audit also found that vasopressor features could include treatment started after intubation, inflating apparent model performance. These preliminary findings suggest that both patient subgroup and assessment time matter when evaluating risk. Identifying clinically distinct groups also motivates a key fair AI question: whether overall model performance conceals uneven errors across patient subgroups. Further validation is needed to assess this possibility and determine the clinical relevance of the identified phenotypes.
