A team of U of T Engineering undergraduate students has created an artificial intelligence (AI) tool that makes official course material accessible through a chatbot-style interface, offering a new paradigm in machine-assisted learning.
The large language model (LLM) — affectionately named Alan after mathematician and computer science pioneer Alan Turing — was originally developed as a fourth-year capstone project and is already being piloted in several undergraduate courses.
The idea was developed by Ethan Hugh, Abhishek Madan, Yassine Berrachid and Vaibhav Seshadri (all ECE 2T5). Hugh and Madan drew on their time as teaching assistants, witnessing how students struggled when learning new concepts or studying, while Berrachid and Seshadri brought in professional perspectives from their internships, where they had explored practical applications of emerging AI tools.
“We wanted to better the learning experience for students without designing something that would just spit out an answer,” says Hugh.
“We wanted something that could act like a helper. For example, let’s say a student is having difficulty understanding a topic; having the course material explained or summarised in a different way was what we wanted to provide.”
“Many of our ideas were things we ourselves would have wanted access to as students.”
Building on these shared ideas, the team began exploring technical approaches together. Berrachid and Seshadri had been dabbling with different applications for retrieval-augmented generation, which helped shape the technical direction of the project.
The creation of Alan became the team’s entry into ECE’s fourth-year capstone course. Throughout the project, they were supervised by Professor Salma Emara (ECE), who continued to develop the project even after the course was completed.
In January 2025, Alan was piloted in APS 105 Computer Fundamentals, a first-year course with around 450 students. At present, the tool has been used in a dozen additional courses with plans for more in the fall of 2026.
“It’s taken off in ways I could not have predicted when the team first approached me,” says Emara.
“In APS 105, for example, we observed clear peaks in usage before deadlines and exams, highlighting the demand for immediate academic support during high-stress periods.”
Alan uses a retrieval-based approach to pull from course content — such as that posted on Quercus, U of T’s online teaching and learning environment — and generate responses.
Because Alan draws directly from course materials, it uses the same terminology and notation as the course instructors. Keeping things consistent has helped reduce student confusion and made it easier to connect answers back to lectures and assignments.
When setting up a course in Alan, instructors can choose from three guardrail modes. They can configure Alan to answer only conceptual and course logistics questions, to provide hints or guidance for analytical or coding questions, or to offer structured, step-by-step solutions to any course-related question.
Unlike existing LLMs, Alan only answers questions within the scope of the course it’s been set up for. It’s also been instructed to say “I don’t know” so as not to mislead users.
Users have access to a like and dislike feature when they get an answer, which has helped provide the team with feedback to inform the tool’s development.
The team is clear that Alan is not there to take on the work of a teaching assistant or professor, but that it instead provides knowledge when teaching teams aren’t readily available.
“Whether we like it or not, students are using AI, but what we don’t want are students to cognitively offload onto AI just to get a final answer,” says Professor Fatemeh Jazinizadeh (MIE), who joined the project shortly after Alan was first deployed.
“We want them to use AI to learn better, more efficiently and to be able to access help 24/7. Neither professors nor TAs are around after midnight to answer student questions, and interestingly, that’s when we know a lot of students are using Alan.”
Jazinizadeh adds that Alan is also making learning more accessible for students who, for a variety of reasons, may not be as comfortable going to office hours for help.
When the capstone team was about to graduate after Alan’s first semester of use, Emara and Jazinizadeh hired Vir Patel, Ada Selcuk and Warrick Tsui (all Year 3 CompE) as research assistants to continue to work on Alan and expand its capabilities.
The research assistants have since introduced multimodal functionality, enabled image-based queries and retrieval of course images, integrated course discussion boards into the knowledge base and migrated the platform to one that requires a U of T login to make Alan more secure. With their work, professors can now set up their course in Alan in under an hour.
Other guardrails have also been put into place over the past year.
“We use prompting to ensure that Alan really stays within the context of the course,” says Selcuk.
“For example, if a student comes to Alan for mental health support, Alan is trained to only provide them with the list of resources from their syllabus. We try to support the students in the best way that we can through these inhibitors.”
Working on Alan has led to new academic and professional opportunities. The capstone team recently published a Special Interest Group on Computer Science Education 2026 poster summarizing early observations of Alan, and both Hugh and Patel credit their research work with helping them land a full-time job and PEY Co-op position respectively.
The project is currently supported through faculty funding, enabling continued development and expansion. The team is also growing, with two new interns joining this summer to support ongoing work on the platform. Alan is continuing to expand across U of T Engineering and the team hopes it will be adopted by other departments and faculties across the University.
Emara says that while commercializing the platform is something they’re thinking about, they have many considerations to address first.
“Ensuring we continue to protect the privacy and security of both students and the instructors’ intellectual property is first priority,” says Emara.
“We wanted to start by proofing the concept, doing market research and making sure Alan is really needed. We now know the need is there and the next step is to try to grow it.”
“At the end of the day, our main goal has always been to help as many students as possible and that’s what we will keep striving towards,” says Madan.
