Building AI tutors for student learning

A chatbot that knows the course, follows the teacher, and keeps a record of what each student has and hasn’t grasped

Dr Jon Cardoso-Silva
📧 J.Cardoso-Silva@lse.ac.uk

Department of Methodology, LSE

30 Sep 2026

Are pedagogical AI tutors delivering?

Brave new words: how AI will revolutionise education, with Salman Khan · DSI Event · Wed 26 Jun 2024

📕 Read my book review.

• Khanmigo was the best bet so far: the most funded tutor, Socratic by design, grounded in Khan Academy, with major philanthropy and Microsoft funding.

• At the DSI event in June 2024, Sal Khan called AI tutoring the biggest positive transformation education has ever seen. By April 2026 he told Chalkbeat that for many students it had been a non-event (so far?).

Dan Meyer, 15 Apr 2026 · RIP Khanmigo & Edtech Industry Dreams of AI Tutors

I reviewed Sal Khan’s book in 2024, when he visited DSI. A Socratic bot will still ask about topics the course introduces weeks later, and it will hand a student a justification they never asked for.

Lessons from prior experience

• Default “Learning mode” / Socratic mode in use by all major AI companies is NOT a good pedagogical feature.
• GenAI seems most beneficial to learning if students take the initiative (self-agency).
• The performance paradox: chatbots give students an illusion of competence, and without process tracking we cannot really know what the student knows anymore.
• Our GENIAL framework, developed from 2023 to 2025, allows us to look at how students actually use GenAI in their self-study and measure these claims.

Cardoso-Silva, Sallai, Kearney, Panero & Barreto (2025), the GENIAL framework, SSRN.

What I’ve been building

• The course convenor sets the story and the calendar, and the tutor follows it.
• The tutor knows which skills are active, which assignment phase students are in, and what to withhold until the course has introduced it.
• Tested on DS105A and DS105W (data science and programming courses), and adapted internally at LSE to MY202A (machine learning), MG317 (organisational learning), AC102 and AC105 (introductory accounting courses), amongst others.

DS105W (Winter 2025/26): formative practice W01 to W04, then Mini Projects across the term.

When to route to what

• The system picks the right response mode for each student message (similar to Claude’s Skills feature) and tags it with an emoji so I can later check which rule was active.

Socratic when it should be

Points back to the W03 Lab, then asks what approach the student will use.

Direct when it should be

The student forgot how to clone a repository, and the tutor gives the steps.

Knowledge tracing

What I am testing now: give the chatbot the skills we expect each student to demonstrate awareness of or proficiency in each week.

Automatic tracking of skills level

The tutor keeps a skills record for each student in its project memory, and it rewrites a row when it sees evidence.

MY202A tutor, tested with me playing the student, 28 Sep 2026. Red: the student’s self-rating before the course. Green: what the tutor recorded after one exchange about lapply().

Each student sees their own skills record

The student gets to see their learning profile.

The tutor turns that record into a study plan

• Currently this is all reactive because we have not built the infrastructure yet, but my goal is to make this part of our VLE (Moodle) and more integrated into the course.
• I also plan to use nudges so that the chatbot becomes more proactive, especially for disciplines where frequent regular practice is beneficial.
• This way we can deliver a scalable version of personalised learning while also delivering dynamic reports to the course leader.

Thank you

Cardoso-Silva, J., Sallai, D., Kearney, C., Panero, F., & Barreto, M. E. (2025). Mapping Student-GenAI Interactions onto Experiential Learning: The GENIAL Framework. SSRN Electronic Journal. (Under Review)

Sallai, D., Cardoso-Silva, J., et al. (2024). Approach Generative AI Tools Proactively or Risk Bypassing the Learning Process in Higher Education. LSE Public Policy Review, 3(3), 7.

Coming this academic year

A paper on the tutor’s design and what we observed (with Dori Sallai and Steven Williams) · A guide for building a chatbot for your own course (Steven Williams, LSE Digital Skills Lab) · A short video series with the Eden Centre

The tutor is for the hours students spend studying on their own. The classroom stays the social space, with the teacher, the GTAs, and each other.

Blog: jonjoncardoso.github.io/blog

📧 J.Cardoso-Silva@lse.ac.uk

The tutor is for the hours students spend studying on their own. When they are in the room with Stuart, the GTAs, and each other, that is where the learning happens socially.