Creighton students learn AI from the inside out
Most people meet generative artificial intelligence the same way: through a screen, a prompt and an answer. It feels like magic. It also feels like something you will never fully understand.
At Creighton, Steven Fernandes, PhD, and Alison Kleffner, PhD, are inviting students behind the curtain.
Fernandes, associate professor of computer science, and Kleffner, assistant professor of mathematics, are helping build a new kind of AI learning environment, one in which students aren’t just using generative AI, they’re learning how to build and deploy applications using it. Thanks to a National Science Foundation research award supporting the project, they can train and deploy advanced AI models without the financial barrier that usually comes with it.
“Building AI models requires Graphics Processing Units (GPUs). These GPUs analyze large datasets quickly but can cost tens of thousands of dollars,” Fernandes says. “The Jetstream2 platform research credits provided to us through the National Artificial Intelligence Research Resource (NAIRR) classroom pilot allows students to build AI models without having to worry about the GPU compute needed.”
It may sound technical, but the result is simple: access.
From Users to Builders
For many programs across the country, the price of serious generative AI work is steep. Training large models often requires expensive computing resources, subscription tools or hardware that students simply don’t have. With NAIRR support, Creighton students can do the work that matters most in the AI economy: experiment, build, test and deploy.
They learn by building projects that prove it.
The NAIRR-awarded project, called “Building Generative AI Applications,” began with a simple question: What would it look like if students could learn generative AI from the inside out?
“With recent advances in AI, we wanted our computer science and data science students to learn how to build and deploy generative AI applications,” Fernandes says. “This motivated us to create a course on this topic.”
That course, CSC 590: Building Generative AI Applications, is a class where students complete hands-on coding. Students build three different generative AI applications — not simulations, but actual applications they build and deploy.
One project focuses on generating synthetic medical images using Generative Adversarial Networks (GANs), a type of AI model used to create new images based on patterns learned from data. Fernandes says synthetic imaging has real potential in medicine, including applications related to early disease diagnosis, because obtaining medical images for training these models can be challenging.
Retrieval-augmented generation (RAG) is being used in another project to build chatbots that respond only using a trained knowledge base, meaning they don’t “hallucinate” information outside what they’ve been trained on. “You can think of it like a virtual tutor that knows your course material and can help you study,” Fernandes says.
Students are also building AI agents, including an AI-powered health coach that can provide personalized guidance for wellness. “In this project, we are not just using existing AI,” Fernandes says. “Instead, we are building AI applications that can assist individuals from medical professionals to college students.”
That distinction matters. In a world flooded with AI tools, Creighton students are learning to build them responsibly. “The goal is to empower students to create AI tools that solve real problems rather than simply consuming AI products built by others,” Fernandes says.
The Math Behind the Models
A project this ambitious doesn’t happen in a silo, and Fernandes is quick to emphasize his collaboration with Kleffner.
Kleffner’s role highlights an important truth about AI: breakthroughs don’t come from coding alone. Strong AI models depend on statistics, modeling and mathematical reasoning; the foundation Kleffner helps students build before they write machine-learning code. “Building AI models requires in-depth statistics knowledge,” Fernandes says, pointing to how Kleffner’s course, MTH 362: Statistical Modeling, strengthens preparation for generative AI work.
One exciting part of the project, Fernandes says, has been watching students go beyond the assignment and into real-world projects.
For instance, student teams created a RAG-based chatbot trained on class material, a practical tool that helps students review content and prepare for finals. Another team developed a local chatbot designed to help new students navigate majors, minors, courses and faculty. These aren’t flashy demos. They are thoughtful applications built by students who are learning to connect AI power to human needs.
A Skill Set for Any Field
Fernandes believes Creighton’s approach extends beyond computer science.
“Students across disciplines who understand how these AI tools work, their capabilities and their limitations will be better equipped to apply AI responsibly,” he says. “Additionally, students who can build and deploy generative AI applications will have a significant edge over those who are just using them.”
In other words, the project prepares students not just for the next job market but for the leadership decisions AI will demand across industries, from healthcare and education to business and science.
