Moving with the course: designing a time aware AI tutor for data science teaching

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

09 Jul 2026

Are pedagogical AI tutors delivering?

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

đź“• 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

LSE AI and Education Fellowships

  • I am one of ten LSE AI and Education Fellows (2025–2027).
  • The fellowship buys out a term of teaching so we can redesign our courses with GenAI: teaching, learning, and assessment.

LSE AI and Education Fellowships · Lourdes Hernandez Martin, Language Centre

How the tutor is built

  • Five prompt files in a Claude Project (all via prompting).
  • Each week I edit CURRENT WEEK and CURRENT PHASE in system-prompt.md. Clock inference was unreliable.
  • weekly/ and assignments/ encode what the course has covered so far.

When to route to what

  • Routing is instructions in protocols.md, not code hooks.

The challenge feature: deeply contextual

The challenge feature: live feedback

Socratic routing in practice

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

Direct support in practice

Student forgot how to clone. The tutor gives the steps.

It reinforces what the course leader said

It aligns with the course leader’s intent for the course.
(The goal should be for “AI in the loop”, NOT “human-in-the-loop”)

But when it goes wrong… it’s VERY annoying

The answer can sound fluent and confident.
It makes you wonder, what other lines has this tool crossed?

The student should be the one justifying its choices. Here, the AI tutor offers a justification when the student didn’t even ask for it!

AI usage always increases near a deadline

DS105W (Data for Data Science) · Winter 2025/26

  • 103 students enrolled
  • 83 agreed to share chatlogs for research
  • 47 shared at least one Claude chatlog

Usage clustered around coursework deadlines.

Preliminary

Caveat: Claude logs only. I still need to verify each chatlog came from inside the course tutor Project.

What I do not know yet

  • I like to think my marking process is AI resilient, but submissions looked particularly strong this year.
    • Is it just performance? Or is it performance with learning?
    • Did students learn more because they had to engage more with the tool? Or did I build a glorified checklist?
  • GENIAL protocol annotation is in progress: emoji markers are an imperfect measure of what the tutor was doing.

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.

Blog series: jonjoncardoso.github.io/blog

📧 J.Cardoso-Silva@lse.ac.uk

Upcoming

(In preparation)