Students use AI for their assignments—this AI tutor can actually help them learn
Researchers created an AI tutor that is trained on course-specific materials and trained never to just give students the answer to a problem
image: San Diego researchers received a grant from the State of California to improve and expand a bespoke AI tutor, stemming from a prototype developed at UC San Diego. Here two students use a pilot version of the tutor to study and do homework.
Credit: David Baillot/University of California San Diego
Video demonstrations:
AI tutor for an introductory programming class: https://youtu.be/K_p5TzBT-kw?si=DEVtUb_2py26ASqU
AI tutor for an introductor nanoengineering class: https://youtu.be/8Ls26PiMRAs?si=gI-jnrS2K76l-F35
A team of researchers at the University of California San Diego developed an AI tutor designed to give students an alternative to off-the-shelf AI tools, so that students not only get help but actually learn course-relevant information at the same time. The hope is that students will find this experience more fulfilling than simply relying on large language models like Google Copilot or ChatGPT.
“The reality is that students will use AI for their assignments,” said Mohan Paturi, a professor in the Department of Computer Science and Engineering at the UC San Diego Jacobs School of Engineering, an affiliate of the UC San Diego Qualcomm Institute and one of the lead researchers on the project. “But those tools do not necessarily facilitate learning.”
The AI tutor is based on an off-the-shelf large language model, similar to ChatGPT. But it is then trained on materials—notes, podcasts and more—for the specific courses where it is deployed. Most importantly, it is trained to never give students the answers to a problem. Instead, the tutor asks questions that lead students to the right answer and encourages them when they do get it right.
For now, the tutor has been available to more than 650 students in computer science and nanoengineering courses at UC San Diego as part of an ongoing pilot program.
The tool was developed by the Laboratory for Emerging Intelligence launched in 2024 in UC San Diego’s computer science department.
The researchers note that their goal is to augment the support provided by professors and TAs, not replace them. One advantage, for example, is that the AI tutor is available 24/7, while human tutors are not. The platform is open source, allowing faculty members to train the tutor for specific courses and related materials.
In fact, the research team is in close contact with the instructors and TAs for the pilot courses where the system is in use, to make sure the AI tutor meets students’ learning needs while following the instructor’s teaching philosophy.
“Our goal is to create a tutor that is available any time and anywhere,” Paturi said. “Long term, we want to build an ecosystem of open-source tutors that other instructors can customize to meet course-specific needs.”
A $1.5 million grant from the State of California’s AI Grand Challenge program will allow researchers to expand the program to eight courses across higher education institutions in San Diego County.
The AI tutor in action: nanoengineering and computer science
Seed work on the AI tutor was supported by a state-funded workforce development program at the UC San Diego Qualcomm Institute, and by seed funding from the department of Computer Science and Engineering. Enabled by this initial support, Paturi was able to work with a large group of interns. One of Paturi's projects focused on developing a large language model (LLM) based AI-tutor called SmartLearning Hub to provide high-quality instructional help.
The AI tutor was first tested in NANO 11, Introduction to Nanoengineering. The introductory class is designed to familiarize students with widely different backgrounds and levels of experience with the fundamentals and applications of nanoengineering and to prepare them for the rest of the courses in the major.
The tutor allows instructors to support students at all different levels of knowledge and meet them where they are at, said Ph.D. student Robert Ramji, who trained the AI tutor and co-authored a textbook for the class with nanoengineering Professor Darren Lipomi, now at the University of Rochester. “We have a responsibility to serve all the members of our student body equitably,” he said.
Lipomi taught the class, where nanoengineering students used the tutor both for reading assignments and problem sets. For example, students could ask the tutor to clarify points in the reading, elaborate or offer further examples, as they read along, in the same browser window. The tutor also provided sample problems—see our video demonstration—and helped students to work through those problems at their own pace. Students also had the option of submitting a short bio to the tutor, so it could further tailor its answers to their interests, such as chemical engineering or materials science.
The students did so well with support from the AI tutor that Lipomi and Ramji considered making the next iteration of the class more challenging.
The AI tutor was subsequently deployed in CSE8, an introductory programming class taught by Sorin Lerner, professor and chair of the UC San Diego Department of Computer Science and Engineering. Half of the students in the course are not computer science majors.
“Programming can be a challenge for novices, so extra support during programming assignments is key,” said Lisa Huang, computer science PhD student and a teaching assistant for CSE 8A.
To get support during a programming assignment, a student types a question into a dialog box for the tutor, which is embedded within the student’s browser. The tutor knows what the final output needs to look like and can see where the code is going wrong. It then guides the student through course materials that can help them to code correctly. You can see an example of an interaction in this video.
Student perceptions
Huang and Ramji said that feedback has been largely positive. Nearly 70 percent of students who took the nanoengineering and computer science courses, where the AI tutor was deployed, described it as an effective or highly effective learning tool.
Some quotes in student evaluations show why they are enthusiastic:
“I like how [the] AI tutor did not give me the answer immediately, instead it gave me tips to improve my code to get to the answer so it really made me learn.”
“[The AI tutor] allows me to take mental and physical notes on the material and learn from my mistakes and feel accomplished after finishing assignments.”
Some students mentioned that they preferred asking the AI tutor because they would be embarrassed asking a person.
A few students did have some critiques. Some were frustrated that the AI tutor didn’t give away the answer; and some preferred in-person help, which they are still able to seek out.
Student perceptions are just one of the tools that the team is using to assess the AI tutor’s effectiveness. The funding from the State of California’s AI Grand Challenge program will allow them to deploy a wider range of tools and to broaden access to the AI tutor platform at UC San Diego, San Diego State University and several community colleges in San Diego County during the 2025-26 academic year.
A laboratory for advancing AI
The UC San Diego AI tutor is part of a broader initiative, the Laboratory for Emerging Intelligence, that works to develop and deploy next-generation AI systems to advance applications in science, medicine, business and education. LEI is led by computer science professor Paturi, in collaboration with Leon Bergen, a faculty member in the Department of Linguistics. LEI faculty also hail from the School for Global Policy and Strategy and UC San Diego Health, in addition to computer science and engineering.