Student Projects

On this page you can find our offerings for Master’s Projects and Master and Bachelor Research Projects in the realm of data mining and machine learning for (vocational) education ecosystems for the spring semester 2026. Please note that this list of projects is not exhaustive and will be updated over the coming days with further exciting projects.

Last update: 21.11.2025

How to apply

Please apply via our student project application form. You will need to specify which project(s) you are interested in, why you are interested and if you have any relevant experience in this area. To access the form, you need to log in with your EPFL email address. If you would like to receive more information on our research, do not hesitate to contact us! Students who are interested in doing a project are encouraged to have a look at the Thesis & Project Guidelines, where you will gain an understanding about what can be expected of us and what we expect from students.

The student project application form will remain open for submissions until the late deadline of 31 January 2026. Applications will be reviewed on a on-going basis, so we strongly encourage you to apply as soon as possible. Early applicants will be considered starting the in calendar week 49, with the early deadline set for 02 December 2025:

• Early deadline (first round): 02.12.2025
• First contact with supervisors: 04.12.2025 – 12.12.2025
• Late deadline: 31.01.2026

External students: Non EPFL students are kindly requested to get in touch with the project(s) supervisors by e-mail.

Please, note that the list of topics below is not exclusive. In case you have other ideas or proposals, you are welcome to contact a senior member of the lab and talk about possibilities for a tailor-made topic.

Project 1: Behavioural and Confidence Profiles in Interactive Science Simulations

Project 2: Joint Semantic Modeling of Information Sources and Student Writing Processes

Producing a coherent written product from multiple information sources is cognitively demanding, particularly for younger learners. Understanding how students navigate sources, take notes, and iteratively revise their texts offers a window into their self-regulated learning (SRL) processes. Despite growing interest in SRL analytics, the computational modeling of relationships between learners’ text edits, their navigation behavior, and the content of the materials they consult remains underdeveloped.
This ML4ED project investigates multimodal trace data from a project-based learning unit in primary school. The dataset includes over 600 students’ evolving magazine articles, their notes, navigation logs, and the 100 text and video resources they consulted. Your task is to design and evaluate methods that connect what students wrote, when they wrote it, and which information sources they accessed. The goal is to predict external SRL ratings of student from this perspective on their learning process. You will work with techniques from Natural Language Processing, representation learning, and sequence analysis. Depending on your chosen direction, the project may also involve large vision–language models or multimodal embedding techniques.

The project will focus on:

• Semantic and structural alignment between sources and writings

• Modeling temporal writing behavior

• Predicting SRL labels from the joint modeling of text semantics and student learning actions

Requirements:

• Proficiency in: Python, NLP

• Interest in: Learning Sciences, Educational Data Mining

• Level: Master

Supervision: Dominik Glandorf (PhD student)

Project assigned: Project 3: Aligning VLM behavior with human-like exploration patterns

Recent advances in vision-language models (VLMs) have enabled them to operate within interactive, open-ended simulation environments. However, little is known about how these models explore, learn, and adapt in such environments compared to humans. This project aims to systematically analyze VLM behavior in open-ended learning simulations.
We will evaluate how multimodal models interact with and navigate simulated environments, characterizing their decision-making patterns, exploration trajectories, and learning dynamics. These behavioral patterns will be compared to human exploration data to identify similarities between artificial and human learners.
Building on these insights, the project will develop methods to align VLM behavior with human-like exploration patterns. By leveraging human exploration sequences and behavioral priors, we aim to create VLMs that can mimic key aspects of human learning behavior in order to generate synthetic exploration traces of human learners.

Required Skills: Experience with VLMs
Proficiency in: Python; ML/NLP libraries
Bonus: Research experience in NLP
Further details:
This semester project is intended for one MSc student (semester project or thesis) with technical background in NLP.  The student will be supervised by PhD student Marta Knežević.

Project 4: Understanding and Diagnosing Faithfulness in Feedback-Generating Language Models Using Sparse Autoencoders

Language models (LLMs) are increasingly used to give students personalised, structured feedback. However, smaller open-source models are still prone to making mistakes: they often miss key errors in a student’s solution, offer irrelevant next steps, or mix up different types of advice. As a result, the feedback they produce is not always faithful to what the student actually wrote.Recent interpretability methods, especially Sparse Autoencoders (SAEs), let us look inside LLMs and examine which internal features activate when they generate different parts of feedback. This project investigates whether these tools can help explain why small LLMs produce unfaithful feedback, and which internal patterns are linked to helpful or unhelpful recommendations.

Objectives:

• Use pretrained SAEs and interpretability tools (e.g., Neuronpedia, circuit-tracer) to analyse how a small LLM processes student answers and generates feedback.
• Identify internal features that correlate with:
  • accurate vs inaccurate detection of student mistakes
  • faithful vs unfaithful next-step recommendations
• Build simple diagnostic tools (e.g., lightweight classifiers or clustering scripts) that automatically flag likely feedback issues based on model activations.
• Develop a reusable analysis toolkit to support future research on improving the faithfulness and transparency of LLM-generated feedback.

• Interest in: Interpretability, NLP, Educational Technology, Model Behaviour Analysis
• Skills: Python, basic machine learning, familiarity with Hugging Face and PyTorch

Project 5: Soft-Skills and Diagnostic Reasoning in a Virtual Pharmacy (PharmaSim VR)

• Design a VR pharmacy environment with multiple virtual customers waiting in line.• Implement varied game character personalities (e.g., impatient, anxious, annoyed, calm) and behaviors.
• Use LLM’s to generate natural, adaptive client dialogue.
• Log interactions for later analysis.Requirements:
• Interest in: Educational technology, VR, Games, Simulations
• Skills: Unity (C#), NLP, LLM
• Bonus: Experience with VR development, conversational agents