About
I am currently a Postdoctoral Fellow at the Department of Human Geography, Lund University, where I also initiated and manage the Hägerstrand Lab , a computational research infrastructure supporting AI-driven and data-intensive projects within the department. My work sits at the intersection of data science and social science, applying machine learning methods to foundational social science questions. My postdoctoral work focuses on using large-scale administrative register data and deep learning to study labour market dynamics, human capital, and mobility. I am particularly interested in developing machine learning models that create meaningful (ideally interventional) latent representations that allow out-of-distribution generalization, transportability from study populations to target populations, complex measurement development, and data simulation. Part of this work is also supported through funding from AI Lund (see my recent AI Lund Lunch Seminar).
My academic interest is guided by the curiosity to understand human systems by combining rigorous scientific design with modern AI methods. I aim to build interpretable, causal, socially actionable models of human systems.
Before joining Lund University, I completed my doctorate in Computer Science (Dr. rer. nat.) at the University of Tübingen in 2024, where I was based at the Hector Research Institute of Education Sciences and Psychology. My dissertation examined cognitive processing in virtual reality learning environments, combining signal processing, machine learning, and experimental methods.
The formal foundation and launch of the Hägerstrand Lab at Lund University's Department of Human Geography.
Co-organising an EU-SPRI autumn school on science, technology, and innovation policy research.
Research Keywords
Machine Learning· Computational Social Science · Representation Learning · Labour Markets · Causal ML · Digital Learning Enviroments · Explainable AI
The Hägerstrand Lab
A shared intellectual and material space bringing diverse strands of computational research in human geography into meaningful proximity.
Documents
Research
My overarching goal is to understand humans through computational, probabilistic, and complexity-oriented science — connecting micro-level cognitive processes with macro-level social dynamics. My work spans two career phases, now increasingly unified into a single computational framework for modelling human-driven systems.
Human Capital & Career Trajectories Current
Within the NEXUS project — Machine Learning on Register Data — I develop latent representation models of human capital using large-scale Swedish register data (LISA, RAMS, CFAR). Individual careers are encoded with transformer-based sequence models to learn low-dimensional embeddings of skills and career states; a first paper was accepted at PAKDD 2026. A parallel stream addresses temporal harmonization of Swedish industry classifications (with A. Erlström, A. von Borries, M. Grillitsch, O. Hall, A. Sopasakis), producing a latent industry representation currently under preparation for publication. Ongoing work integrates graph-based employer–employee relations and firm transitions to analyse career mobility as a relational process. Together with Yuan Liao and Markus Grillitsch, I also recently started a project combining mobile phone and register data to study inter-firm knowledge connectivity and labour mobility as mechanisms of regional learning and economic transformation.
Opportunity Structures & Spatial Inequality Current
This strand examines regional opportunity structures, industrial change, and spatial inequality. A central project develops latent industry representations for studying regional path development — constructing a shared representation space that combines worker mobility, semantic similarity, hierarchical industry structures, and input–output relations. Building on this, the concept of an opportunity space links latent industry structures with region-specific economic compositions, allowing regional development, industry emergence, and industrial decline to be analysed as geometric and relational processes. A complementary project with Lorena Melgaço, Luana X. P. Coelho, and Fabio Schreiber uses three decades of grid-based census data to study racial spatial inequality and urban transformation processes in Brazil, examining how racialized spatial inequalities persist beyond socioeconomic differences.
AI for Society: Vision Models & Media Analysis Current
I apply computer vision and large language models to socially grounded questions. Together with Ola Hall, Eva-Charlotte Ekström, and Magnus Jirström (in cooperation with researchers in Addis Ababa), one project applies vision–language models to analyse advertising practices in Ethiopian souk shops — investigating how visual communication strategies relate to local economic conditions, market structures, and informal institutional contexts. A parallel project with Jonathan Friedrich, Melissa Cardona, and Linda Stihl examines institutional adaptation and policy failure in EU textile waste governance through AI-assisted newspaper analysis, tracing discursive shifts and institutional coordination failures across large-scale corpora. This project also serves as a methodological testing ground for integrating LLM-based approaches into theoretically grounded social-scientific text analysis. A third strand uses semantic identification frameworks to study informal institutions in green transition policy texts.
Causal Machine Learning Emerging
An emerging strand in my agenda is the move from predictive to causal inference in computational social science. Much of contemporary machine learning excels at prediction but remains limited in identifying underlying mechanisms, intervention effects, and transferable causal structures. I am therefore developing causal representation learning and structural causal model approaches that bridge representation learning with interventional inference — producing models that are explanatory and policy-relevant, not just predictively accurate. A concrete application under development (with Ola Hall, Alexandros Sopasakis, and Marina Toger) applies machine learning and satellite imagery to develop longitudinal uncertainty-aware imputation methods for studying the relationship between temperature and health, submitted to the Swedish National Space Agency (Rymdstyrelsen). As a first community-building step, I am planning a workshop on causal machine learning with observational data.
Virtual Reality, Eye Tracking & Cognition PhD Research
My doctoral work established VR as an experimental platform for studying learning and cognition. I developed immersive VR classrooms and multimodal data pipelines showing how presentation modes affect attention, mental rotation, memory consolidation, and learning outcomes. A key methodological contribution is the use of eye tracking as a proxy for latent cognitive processes: gaze transition entropy, attention network analyses, and interpretable ML pipelines to infer cognitive load and teaching expertise — published in Scientific Reports, Computers in Human Behavior, IEEE VR, ISMAR, and ETRA.
Language, Interaction & Teaching Quality PhD Research
Using large language models and text embeddings to analyse classroom transcripts, I demonstrated how AI can scale qualitative educational research while maintaining interpretability — including GPT-assisted rubric-based scoring and NLP-based interpretability with SHAP values. This work bridges my cognitive science background with my current interests in AI-assisted social science.
Publications
Full list on Google Scholar.
Teaching
Student-Centred Learning
Students are co-producers of knowledge. Interactive activities, group work, and problem-based exercises replace one-directional transmission.
Formative Feedback
Short feedback loops — weekly forms, iterative reflection — help students monitor their own progress and adjust learning strategies.
Reducing Cognitive Barriers
For quantitative content I use clear structure, step-by-step scaffolding, and explicit management of statistics anxiety.
Technology & AI Integration
Browser-based coding (Google Colab), flipped video inputs, and critical discussions of AI tools are integrated throughout courses.
723.5 hours total. Full details in the Teaching Portfolio.
Educational Material
Interactive teaching materials combining visualisations, mathematical derivations, and embedded code exercises.
Contact
223 62 Lund, Sweden