Bar-Ilan aims to bring its AI study agent to 100 more courses

Bar-Ilan University plans to bring an AI study assistant to another 100 courses, giving more students a way to search lectures and quiz themselves on assigned material. Microsoft’s October 15 account describes the next stage of a university deployment built around instructors’ own resources, with encouraging student and lecturer experiences rather than proven learning gains.

From borrowed notes to a course assistant

For student Yarden Shay, the problem began with missed classes. Studying criminology and psychology, she struggled to catch up on legal-responsibility lectures while keeping pace with six to eight hours of new lectures each day. Borrowing classmates’ notes left her chasing information she needed to understand the next lesson.

In February, the university piloted an AI agent inside Moodle, its online learning platform. The trial covered 71 courses and about 2,400 students. Students could ask questions in ordinary language, find a topic within a lecture, request a summary or generate a quiz to test their understanding.

Microsoft describes the agent as grounded exclusively in materials uploaded by teaching staff: recorded lectures, transcripts, presentations and articles. That confines its intended source material to the course itself. KPMG and OpenApp developed it using Microsoft Copilot Studio and Azure AI Services. It runs in the university’s Azure environment and supports Hebrew, Arabic and English.

The lecturer had to teach the bot first

Making those resources usable required work from lecturer Eliyahu Biton, who helped test the agent before it went live. For his online healthcare law and ethics courses, he revised materials, supplied transcripts of every video lecture and supplemented slide bullet points with research papers and recommended reading.

Biton then tested the responses until he was satisfied with their accuracy. His account makes the preparation concrete: uploading existing slides was not the end of the job. He expanded the material so the agent could explain what he wanted students to learn, rather than simply retrieve terse teaching notes.

If I want it to be my assistant, I need to teach it.

Eliyahu Biton, Bar-Ilan lecturer, speaking to Microsoft Source

Early benefits, without a verdict on learning

Shay now uses the agent for active study, not just catching up. She asks for multiple-choice and open-ended questions, then gets feedback explaining why an answer is right or wrong. She told Microsoft she spends about three hours studying after lectures each day, down from five to six hours previously.

Biton says he receives fewer routine and administrative questions, and that students seem less nervous about exams. Instead of relying only on old exam papers, they can ask the agent to generate additional practice tests. Those are his observations of how students prepare, not a measured result across the university.

At the August finals, his roughly 50 students averaged three percentage points more than students in previous years. Biton acknowledged that this could reflect a stronger cohort. The comparison does not establish that the agent caused the improvement; likewise, Shay’s shorter study days describe one student’s experience rather than a university-wide outcome.

The next rollout is planned for the first semester of 2026/27, which began in October. Meanwhile, Biton still holds 90-minute discussions with students four to six times a year. He sees his role shifting toward mentoring: helping students learn and approach real-world challenges, while the agent makes course information easier to find.