Practical AI in Action at the U of A’s Graduate College

When the University of Arizona Graduate College began exploring ways to better support prospective students, current students, and graduate coordinators, Patrick Barabe approached the topic with caution.

As IT Director for the Graduate College, Barabe has spent years evaluating technology solutions and vendor claims. Like many experienced technology leaders, he has learned to approach emerging tools with a healthy degree of skepticism.

“Marketing teams will make it all sound super awesome,” Barabe said with a laugh in a recent interview. “In my experience, it’s never as good as it’s cracked up to be.”

Yet less than a year later, the Graduate College’s AI-powered virtual assistant has become a practical, widely used support tool that helps connect students and campus partners to the information they need, often in just one or two questions.

The original goal was straightforward: provide a channel for general questions to be answered quickly and to reduce the burden on frontline staff who spend significant time answering repetitive questions by email and phone. The Graduate College supports prospective students, applicants, current graduate students, graduate coordinators, and directors of graduate studies across dozens of academic programs.

“We were considering ways to supplement or complement the efforts of frontline staff,” Barabe explained. “We wanted a tool on our website to help people find the information they’re looking for.”

Initially, when the Graduate College began exploring several external vendors, he found that many of the available products were too narrowly focused on recruitment and carried price tags that were difficult to justify. At the same time, Barabe was hearing about internal experimentation with AI assistants across campus.

After reaching out to colleagues in UITS and other technology circles, he connected with Jose Fragoso, Assistant Director of IT, who directed him to AI work by Jay Timsina, Enterprise IT Architect, and his UITS team, already in use by Housing & Residential Life. 

Liking what he saw, Barabe and the Graduate College leadership team met with Timsina to learn more about the university-developed solution. “Our budget is definitely limited,” Barabe said. “Supporting an internal university project checked a lot of boxes.”

The project immediately stood out for several reasons:

• It could leverage existing website content as its knowledge source.
• It offered flexibility beyond recruitment-only use cases.
• It could be implemented quickly.
• And perhaps most importantly, it was affordable.

One of the most surprising aspects of Barabe’s implementation experience was how quickly the AI assistant became operational. Once the Graduate College identified its websites as the primary source of knowledge, the implementation process accelerated.

Using automated web scraping tools, the assistant could continuously scan the Graduate College’s web content and build a searchable knowledge base for the AI assistant. “That took very little time,” Barabe said. “Jay was able to give us a test instance to pilot and review within a couple of days. It was less than a week.” The speed stood in sharp contrast to the timelines often associated with external vendors. “No outside vendor could do that,” he said.

The pilot process enabled Barabe and subject matter experts across the Graduate College to test real-world questions that prospective students, current students, and graduate coordinators might ask. The results were encouraging almost immediately.

“Most of the responses we were getting were really good,” Barabe said. But as with any AI implementation, the process required refinement.

One challenge stemmed from the complexity of graduate education terminology. Similar language is often used across admissions, current students, graduation requirements, and departmental policies. The team found that some questions needed clearer contextual guidance. “We figured out ways to adjust the instructions so the AI was better able to determine whether the question was admissions-related, for a prospective student, or for a graduating student,” Barabe explained.

The assistant’s knowledge base updates automatically every six hours by re-scanning the Graduate College website. “That’s one of the things that’s super useful,” he said. “New information is reflected automatically.”

Even after launch, Barabe continues to review conversation logs weekly to identify opportunities for improvement. The logs are anonymized but still provide valuable insight into how users interact with the assistant and where refinements may be needed. Ironically, AI itself became part of that refinement process.

“I try to identify places where the response might have been a little bit better and think about how to tweak the instructions,” he said. At one point, Barabe used Claude AI to help reorganize and improve the instruction set, guiding the assistant. “I copied and pasted the instructions into Claude and asked if it could improve their organization in a way that would be useful to the AI assistant,” he said. “It actually did a really good job.”

One of the Graduate College’s primary goals was never to replace human interaction. Instead, the AI assistant was designed to help users quickly find the right resources and contacts. That distinction has proven important. The Graduate College manages admissions for every graduate program at the university, but individual departments set their own admissions criteria and academic requirements. As a result, many student questions ultimately need to be answered by a specific academic program.

“We wanted to help people get to the right place to ask their questions,” Barabe said. The AI assistant has become especially effective at directing students to department-specific contacts, graduate admissions guides, and program resources. For example, a student asking about prerequisite coursework for a master’s degree in electrical engineering may not need an answer directly from the Graduate College. Instead, the AI assistant can quickly direct them to the department best equipped to help.

“That’s been pretty successful,” Barabe said. Although Barabe still approaches AI thoughtfully and pragmatically, the experience has shifted his perspective. “Certainly for this use case, I’d say I’m now in favor.” He acknowledges there are still limitations. Highly personalized or uniquely complex situations may still require direct human assistance. But he now sees clear value in practical AI applications that enhance the user experience and reduce repetitive work.

The project has also shaped his professional exploration of AI tools. As the lead developer and architect of the Graduate College’s admissions system, Barabe has increasingly experimented with AI-assisted coding and development workflows.

“It can definitely be helpful,” he said. Still, his approach remains grounded. Rather than viewing AI as a replacement for expertise, he sees it as a tool that can extend and support human work when used carefully.

The Graduate College’s virtual assistant may be just one implementation among many emerging AI initiatives across higher education, but Barabe believes these early projects matter. They offer real-world insight into how AI can improve services today, not as a futuristic concept but as a practical operational tool. 

For Barabe, one of the biggest lessons is that successful AI adoption does not necessarily require massive infrastructure investments or lengthy deployment cycles. Sometimes it starts with a clear problem, a collaborative internal partnership, and a willingness to experiment.