The Story — Local-AI Lecture Assistant for UDL

Setting the scene

Dr. Trevor Thang is the course director for a graduate, in-person course taken by every PGY1 resident in the program. Like most lecture-based teaching, it leans on two things happening at once: the slides on the screen and the live voice in the room.

That works beautifully — if you are in the room, on that day, processing information the way the lecture is delivered. Dental education does not always grant those conditions.

The problem

Two gaps kept showing up. First, absence: residents on clinical rotations, or away with illness, miss synchronous content and then struggle to reconcile what they missed against their learning objectives. Second, format: some students process spoken material poorly, some are working in a second language, and some simply learn better from text than from audio.

            The honest limitation: slides are an adjunctive tool. They were never meant to be a comprehensive record of the lecture — so a student who only has the slides is missing most of what was actually said.
        

The solution

Rather than ask students to adapt to a single delivery format, the course adopted the idea of multiple means of representation from Universal Design for Learning (UDL): give the same content in more than one form, and let students choose how they engage with it.

The vehicle for that is a locally-hosted AI lecture assistant. It takes a lecture recording and turns it into a high-fidelity transcript, a concise summary, structured study notes, explicit learning objectives, and practice quiz questions. The lecture is captured once; the learning material comes out in five forms.

            Why local matters: the tool runs entirely on internal University of Toronto infrastructure. Lecture recordings — and anything said in them — never leave the building. Data security and privacy are a design constraint, not an afterthought.
        

Human in the loop

Generated material is never published blind. A "human-in-the-loop" protocol lets the educator review every output for clinical accuracy before students see it, and adjust how the tool behaves — whether to draw on external knowledge bases, and stylistic choices like bulleted versus narrative formats. The AI does the time-intensive drafting; the clinician keeps the rigor.

See how it works →