Student-Designed Customizable AI Platform Earns Education Award
Before winning the Education and Educational Tools Award at the University of California, Davis, Big Bang! Business Competition, CourseAssist AI was a new platform that stemmed from a personal problem for Ty Feng as a TA in "Programming Languages" in the summer of 2023.
"It was a large class, and there were more students than I could help with at a time," said Feng, who recently graduated with his Master of Science degree in computer science from UC Davis. "We had two TAs for over 100 students, so it was a challenge to provide adequate support to every student in the class."
Feng also noticed that the students turned to AI systems like ChatGPT for help. Chatbots are taught from the internet, he points out, and give misleading and incorrect answers because their training is more general and not as specific as it needs to be to help students learn. For many upper division courses especially, there is often a lack of quality information on the internet, so ChatGPT often gets things wrong when asked about these courses.
Students were seeking a helpful AI resource for class, and TAs and instructors at UC Davis needed help offering personalized support for each student.
CourseAssist AI was born. The customizable chatbot is trained solely on a course's materials. An instructor uploads anything pertaining to the course, including a textbook pdf, lecture slides, notes and source code for practice problems. The instructor then clicks a button to train the AI, which takes about three minutes, and it's ready to use. Students are sent a link to their course's personalized chatbot.
Instructors and TAs can also customize CourseAssist AI for how much help a student receives. For instance, it can be programmed to ask guiding questions to point the student toward the answer, but it can also be programmed to provide bullet points for the student. The answers can provide links to the slides and materials the student can review, which can be particularly useful for studying for finals.
Hoping to find other like-minded individuals interested in building the new educational tool, Feng pitched the project to his peers in the UC Davis Prem Jain Computer Science Startup Club, advised by Dipak Ghosal, professor and chair of computer science. Kunal Mundada and Sa Liu, both in the Computer Science Graduate Group, were intrigued by the idea. Mundada, particularly, was drawn to the challenge of the project.
"I did initially think that the problem was quite challenging," said Mundada. "To make large language models, or LLMs, more accurate and appropriate for education, we need to innovate in AI research and development, tailoring LLMs specifically for educational use cases. It was an interesting problem, and if executed well, it could benefit a lot of students and instructors."
CourseAssist AI launched in December of 2023 in the same programming languages class in which Feng was a TA, and within 10 days, the platform had over 70 users. By the winter quarter of 2024, 170 users were utilizing it across three courses: "Operating Systems & System Programming," "Introduction to Programming" and "Ethics in an Age of Technology." This past spring, 300 students were using CourseAssist AI in "Web Programming" and "Introduction to Data Structures" courses.
With overwhelmingly positive feedback coming from students and instructors, Feng, Mundada and Liu, decided to enter the Big Bang! Business Competition.
Organized by the Mike and Renee Child Institute for Innovation and Entrepreneurship, the contest offers workshops and business advice for budding entrepreneurs. At the end, the contest selects a project as the overall winner and categorical winners in topics including Education and Educational Tools, Animal Health and Industry, and Energy and Sustainability.
Earning the Education and Educational Tools Award and the $12,500 prize was surreal, said Feng. However, the experience of learning how to grow a business was really the takeaway for Mundada.
"The feedback and experience we got from pitching to actual venture capitalists, or VCs, was important," he said. "We were able to understand what we should prioritize and deprioritize, and a lot of things got much clearer for us, which enabled us to make more progress."
The team incorporated CourseAssist this past March with Feng in the role of CEO, Mundada as CTO and Liu as a founding engineer. They plan to finish making improvements to the platform's current iteration and begin talking to VCs for seed funding. Their goal is to expand CourseAssist to five universities in the first year.
While the tool is currently tailored toward computer science classes, the group also aims to add features that will make it useful for other STEM disciplines. Introductory courses in STEM are often a student's first deep dive into the subject and have a particularly steep learning curve.
"If they don't have meaningful support in these classes, students might think it's too hard for them and switch majors," Feng said. "Providing that support could make a significant difference in the learning outcomes and could improve student retention and performance in the major."