Open dissertation [PDF] ↗ Thesis Showcase // TU Delft // 2026
Sequences in Data to Represent Learning Processes
Manuel Valle Torre
For a maguey to produce a towering quiote—or to be harvested for mezcal—its growth must happen in the right order and over the right amount of time. If this long, and mostly invisible, process is short-circuited, the output lacks substance. The same is true for learning.
The problem
A polished product can hide an unfinished process.
With the popularization of AI, both a novice delegating all the work and a more experienced hard worker can generate identical results in seconds; the final product is no longer a reliable measure of competence. However, we have the tools to make students' learning processes visible and be able to support them in real time.
Interactive research journey
Follow the growth cycle
Select a phase to trace how the work moves from observing learning sequences to supporting learners while they work.
Phase 01: Analyze
Make the process visible
The first task is to look beyond final products. Sequence analysis can reveal the structure and fine-grained actions of problem solving, but static artifacts still conceal the trial-and-error that produces them.
EdX Log Data Analysis Made Easy: Introducing ELAT ↗
Sequence and process visualization for MOOC logs at scale.
Chapter 3 // LAK 2024The Sequence Matters in Learning ↗
A systematic review identifying the gap between analysis and real-time intervention.
Chapter 4 // arXiv 2026What makes an Expert? ↗
Comparing problem-solving practices in data science notebooks.
Propositions
Ten positions on process, AI, and learning.
- 01
Thanks to Generative AI (GenAI), the quality of unsupervised artifacts--- like an essay or code snippet---went from imperfect to useless as a proxy for student competence. (This thesis)
- 02
Unproductive help-seeking with GenAI does not self-correct over time; without explicit intervention, use of AI tools tends toward dependence rather than learning. (This thesis)
- 03
The value of AI in education lies not in optimizing performance but in supporting the learning process: a well-timed encouragement can help more than a precise explanation. (This thesis)
- 04
Delegation is a valid professional strategy; education must shift from forbidding GenAI delegation to teaching students when to use it and how to formulate and verify the output of that delegation.
- 05
We design GenAI tutors to be conversational partners, but students use them as vending machines; pedagogical design must bridge the gap between how tools are actually used and how we imagine they should be. (This thesis)
- 06
Learning Analytics often falls into analysis paralysis, carefully mapping the learning journey without actually intervening, while GenAI implementation is opportunistic; effective educational technology requires us to find a middle ground.
- 07
The view from the mountain top may be the same whether you arrive by bicycle or by car, but the joy is in the pedaling; in our rush to automate the outcome, we risk losing the pleasure of the process.
- 08
Educational research can no longer afford to be paralyzed by systemic inertia; to keep pace with technological change, schools, teachers, and students must become active participants in the research process.
- 09
Beware of the illusion of competence: plumbing tutorials and instructions will give you the required knowledge, but only water will tell you if your installation is right.
- 10
Writing a dissertation on `process over product' is ironic, given that the work is ultimately judged by what is captured in this book rather than the years of process.
The contribution
From data to learning.
Instead of chasing flawless final products, this dissertation explores how we can use data to uncover the learning process as it happens. It investigates practical ways to deploy real-time conversational interventions, showing how sequence-driven scaffolding can nudge students in the right direction and prevent them from treating AI as a solution vending machine.