Project Duration: January 29, 2023 – May 9, 2023

Role and Responsibilities

In this project, my responsibilities included conducting user research, programming the ChatGPT model with specific knowledge, and setting up APIs to ensure that our chatbot, Topper, provides accurate class recommendations and information for the students of St. Edward’s University.

Overview

For our senior studio project, my team of three other students and I developed an AI-powered chatbot named “Topper.” This chatbot was designed to simplify the academic planning process, especially for elective courses. Topper aims to assist students during times when direct staff support is limited or unavailable. By enhancing the user experience with an intuitive AI tool, we sought to integrate advanced technology into school services.

Problem Statement

Choosing the right courses can be a daunting task, particularly for new students. The absence of personalized guidance and streamlined communication outside of office hours often leaves students relying on emails or scheduling appointments with advisors. These challenges can lead to frustration and disengagement during the enrollment process.

Technology Integration

To address these issues, we utilized OpenAI’s GPT technology, leveraging its playground, fine-tuning capabilities, and assistant API configuration. This allowed us to create natural and effective interactions between students and Topper. One of the major challenges we faced was ensuring that Topper’s responses were accurate and relevant to our specific dataset, avoiding the generation of random or incorrect answers. To overcome this, I developed over 500 unique queries to train the AI, ensuring it could provide precise information. The integration of Topper with the API and our data was key to achieving a cohesive system.

Prototype & User Testing

Our team adopted an iterative design approach, using Figma for prototyping and Miro and Google Sheets to organize our ideas and gather feedback. We conducted two rounds of user testing via Zoom, recording the sessions to capture real-time interactions and feedback. Additionally, we distributed a Google Forms survey to gather more insights. These user tests revealed that students could find courses in under five minutes, allowing us to refine Topper’s functionality and interface based on the feedback received.

Key Features

• Personalized Course Advice: Topper analyzes each student’s academic needs and goals to offer customized course recommendations.
• Conversational UX/UI: We designed the user interface based on common chat interfaces, making it familiar, user-friendly, and accessible.
• Future Integration Goals: Although we cannot integrate this project into the school system due to privacy concerns, our goal is to eventually have Topper seamlessly integrated with existing university academic systems and databases.

Challenges Overcome

Programming Topper to deliver data-specific responses was our most significant challenge. Initial attempts yielded results that were not tailored enough. To address this, we revised our approach, shifting from our initial plan to use Google Cloud and a local server for the API. Instead, we adapted to new updates on the OpenAI API platform, which I learned to use on short notice. This transition, coupled with robust filtering and the use of Python and Flask coding, ensured that Topper’s responses were relevant and reliable.

Conclusion

The development of Topper provided invaluable insights into how AI technology can be leveraged to solve real-world educational challenges. By integrating advanced AI with a deep understanding of student needs, we created a tool that not only improves the course selection process but also serves as a scalable model for future enhancements in academic advising.

Watch our Project Video