Enhanced Learning with AI: A Student-Led Flashcard Generator Project, FHNW School of Business

In a world where education and technology increasingly intersect, a team of students from the FHNW School of Business is exploring new ways to enhance learning. Valerio Job, Stefan Ilic and Kenan Ahmetasevic, students in the bachelor’s programme Business Information Technology, have developed an AI-powered flashcard generator – a tool designed to improve the study process for students.

Throughout his studies, Valerio, like many students, found it challenging to efficiently master complex technical content. Traditional study methods, such as creating flashcards, were helpful but also very time-consuming. While flashcards are known for their effectiveness due to their portability and ease of use, creating them by hand requires a significant amount of time and effort.

Motivated by a desire to simplify this process, the team began exploring how generative AI and large language models (LLMs) could be used to automate the creation of high-quality flashcards. Their goal was to create a tool that could convert lecture slides into ready-to-use flashcards, helping students study more efficiently.

Initially, Stefan and Kenan built the first prototype as part of a study elective. This prototype successfully created flashcards based on lecture PDF slides, using OpenAI’s GPT-4 model integrated through an API. However, the team quickly encountered a significant challenge – data privacy. Since the GPT-4 model is closed-source and hosted on external servers, it was impossible to ensure the privacy of the lecture content.

Driven by a passion for AI and recognising the need for better data privacy, Valerio took on the task of optimising the prototype and turning it into a Minimum Viable Product (MVP). He focused on improving four key areas: Text Extraction, Text File Splitting, Data Preprocessing, and the Model itself.

Valerio turned to open-source models available on the machine learning and data science platform HuggingFace and fine-tuned them using a dataset he carefully curated from different university lecture slides of his study programme at the FHNW.

With these improvements, the team was able to create an MVP, named «flashcardsGPT», which could transform lecture slides into flashcards using open-source LLMs that run locally, offering better data privacy and integration flexibility.

This project garnered attention from Valerio’s professor, Rainer Telesko, who encouraged him to present the work at EDULEARN24, a prestigious conference that brings together educators, researchers, and technologists to discuss the future of education. The conference was held in Palma de Mallorca from 1-3 July 2024 and its theme – «Connecting Technology with Education» – aligned perfectly with the project.

Being accepted as a speaker at EDULEARN24 was a significant opportunity for Valerio, especially as an undergraduate student. Among over 800 attendees, only a few undergraduates were invited to present, making this a notable achievement for him and the team.

At the conference, Valerio shared the potential of flashcardsGPT with an interested audience, demonstrating how the tool could save students and professors valuable time by automatically generating flashcards from lecture materials. This innovation not only makes studying more efficient but also ensures that students have access to high-quality revision materials, which are crucial for retaining complex information.

The project’s focus is currently on flashcards, but the team is also exploring the possibility of expanding the tool to generate other study materials, such as multiple-choice questions. The overall aim is to help professors create various types of study aids quickly, enhancing the learning experience for students.

The response at the conference was positive, with attendees recognising the broader implications of such a tool in educational settings. FlashcardsGPT is an example of how AI can be leveraged to address everyday challenges in education, making learning more accessible and manageable for students.

The journey is still ongoing. Valerio and his team continue to work on improving flashcardsGPT, with plans to refine the model further and expand its capabilities. Future updates may include more advanced optical character recognition tools and the application of retrieval-augmented generation techniques to enhance the quality of the flashcards, as well as implementing further study materials besides flashcards.