AWS Public Sector Blog
How the Wharton School built a virtual TA chat assistant with generative AI on AWS
The Wharton School of the University of Pennsylvania has been at the forefront of business innovation for over 140 years. The school has a long-standing emphasis on the use of data and technology in both its research of business ideas and its approach to student learning. In keeping with this philosophy, Wharton fully migrated its infrastructure to Amazon Web Services (AWS) in 2020 so it could move quickly and adapt to the latest technological advancements.
Recently, Wharton found an opportunity to use generative AI to enhance their students’ learning experience. With support from AWS, Wharton will roll out a virtual teaching assistant (TA) in the autumn of 2024, powered by generative AI, to answer students’ specific course-related questions in real time.
Filling a learning gap at Wharton with generative AI
Message boards are a common way for student learning to continue outside the classroom. When students have questions, they can post on their course’s message board and seek answers from their peers, course TAs, and professors. These message boards are often monitored by TAs who are juggling numerous responsibilities. They can’t always respond to student posts in a timely way, so students can wait hours or even days for an answer to their questions.
This challenge presented an opportunity for Dan Alig, former chief information officer (CIO) for the Wharton School and recently named AWS Champion. He explained, “Part of Wharton’s mission is to incorporate innovative technologies into our curriculum, into our practices.”
In the spring of 2024, Alig and the Wharton computing team determined they could use generative AI to deliver a more accessible learning experience: a virtual TA, available to answer student questions immediately, day or night.
To help develop a proof-of-concept (POC) for the project, Alig reached out to Peter Fader, the Frances and Pei-Yuan Chia Professor of Marketing at Wharton. As an expert in behavioral data analysis of customer shopping and purchase activities, professor Fader has extensive data analysis experience and develops course material based on his many published books. This helps avoid copyright concerns when training the large language models (LLMs) at the core of the virtual TA. Fader was the ideal faculty partner to trial-run the virtual TA project.
Building a secure and dynamic chat-based AI assistant in two months
Alig and his team met with Wharton’s strategic partnership team, which assists faculty in translating their research ideas into practical projects. Quickly grasping the promise of the virtual TA concept, the team connected Alig to an array of AWS resources, including a team of data scientists, AI specialists, and cloud practitioners.
Wharton took a cautious approach going into the project. “When we started this process, we knew that pumping a bunch of data into an LLM and hoping for the best was not an experience that we were willing to put the Wharton name on,” Alig said. “One of the most valuable things that AWS helped us achieve was really preserving the integrity of the project, making sure that it was representative of Wharton.”
Working with AWS, the Wharton team completed the POC in about two months. They used Amazon Bedrock, a fully managed service that lets users build on top of high-performing foundation models (FMs) from leading AI companies, and a variety of data provisioning tools.
Alig credits AWS for helping speed the POC’s development, saying, “The AWS team was instrumental in training and testing the LLMs. Their data scientists and engineers were able to manage how the LLMs were engaging with and consuming the content, and then responding to student conversations.”
This process helped make sure that the virtual TA served only accurate and well-sourced information to students.
Providing new learning opportunities with generative AI
The virtual TA will roll out to Professor Fader’s executive MBA class in fall 2024. The student experience will be straight forward: They can access the virtual TA through the Wharton learning management system (LMS) and, using a basic chat interface, ask the virtual TA course-related questions and receive responses in real time. Not only does the virtual TA answer questions as they come up, but it also presents specific reference material to support its answer—including previously recorded video lectures—so students can dive deeper into the material.
This AI assistant represents another pathway for students to strengthen their understanding of the course material. “Right now, information is becoming available in different modalities and channels, different ways that weren’t available historically,” says Alig. “This is our attempt to lean in and understand how they are going to impact students’ ability to learn material and retain that information.”
From an administrative perspective, Wharton hopes the virtual TA can alleviate some of the message board oversight that burdens their TAs, freeing up their time for more direct and focused student support as it’s needed. Alig hopes the virtual TA can help Wharton gain an understanding of how they might more widely deploy generative AI projects on campus.
“This is the prototype,” Alig said. “We’re going to learn a lot and then we’re going to iterate and grow it. AI is going to have a presence in our curriculum and our ecosystem. This is one of many steps forward, and we’ll evaluate where we go and continue to adapt, based on the technologies available through AWS and our other providers to help empower students.”
Learn more about generative AI for higher education
Colleges and universities are digitally transforming their operations with cost-effective, scalable, secure, and flexible AWS infrastructure. Learn more about how higher education institutions are using AWS at the AWS for Higher Education hub.
At the recent IMAGINE conference for education, state, and local leaders, education leaders joined AWS experts to discuss the future of generative AI in research, higher education, and more. Watch the recorded breakout sessions to discover how schools are building responsible generative AI applications, examining the adoption of generative AI as part of the student experience, and more.
Read related stories on the AWS Public Sector Blog: