Pitt extends AWS-backed AI center, opening student projects beyond campus

The University of Pittsburgh’s AWS-supported AI Cloud Innovation Center is moving beyond its original one-year remit. Now in its second year, it has extended its contract, added students and disciplines, and opened the door to projects from local government and nonprofits. The program pairs practical AI development with training students for cloud-technology jobs.

Pitt described the extension in an October 29 update. The center began with a focus on health sciences and sports analytics; its broader scope now gives students work across more fields. Being open to outside proposals does not mean those organizations already have projects underway.

Tools, funding and a client to answer to

Amazon Web Services supplies free access to its tools, funding for students and an on-campus solutions architect to help with the work. Students and center staff tackle problems brought by faculty and staff, rather than working only through classroom exercises.

That setup requires more than programming. Students meet the people requesting a solution, work through unfamiliar subject matter and present their projects. Kate Ulreich, Pitt Digital’s lead for digital marketing strategy, said the center prioritizes problem-solving, communication skills and drive when choosing students.

We really view this as a workforce development program for the students

Kate Ulreich, Pitt Digital’s lead for digital marketing strategy

An eight-week prototype for a stubborn data problem

One earlier project illustrates the work behind that model. In fall 2025, medical researcher Christopher Horvat brought the center a healthcare-data problem: hospital systems use different codes for information such as lab results and medications. Researchers must translate those codes into a common vocabulary before they can reliably analyze the data.

Gary Farrell, now a computer science and data science junior, took it on as his first center assignment after initial training. He received a spreadsheet containing 40 million rows of codes and built an automated matching prototype in eight weeks, meeting regularly with Horvat and drawing on staff support.

Pitt Digital’s December 10, 2025 project account explains how the EHR Code Mapper works. Developed using synthetic data derived from thousands of real patient encounters, it divides the task into two stages:

1. Narrow the choices. The system compares meanings to select a smaller pool of plausible standard codes, reducing the candidates from thousands to 30.
2. Evaluate the shortlist. A large language model weighs those candidates using associated values or categories and, where available, how frequently each standard code appears in real-world use.

Pitt Digital reported 93% accuracy for exact matches and 98% when considering the top three suggestions. The tool also supplies reasoning and a frequency rank so researchers can validate its suggestions. Those figures describe the prototype; Pitt’s October update says Horvat is still working on deploying it.

The prototype is free and open source, allowing others to use and modify it. Pitt says that public release is standard for the center’s projects, making the resulting software available beyond the researcher or department that originally requested it.

Career success creates a staffing challenge

The workforce mission has a practical complication. Center leader Joe Yun, also Pitt Digital’s chief AI architect, said students’ experience makes them attractive to employers. Competitive jobs and internships can pull them away, making it difficult for the center to retain the people it has trained.

Farrell took a summer internship at PPG in cloud platform engineering, then returned in the fall. He is now on his fourth center project. His assignments have ranged from aligning X-rays for orthopedic surgery to analyzing learning time in Canvas—work that requires learning a client’s field as well as using AWS tools.