Zurich · language, data & music
Computational linguist detecting cultural moments in music streaming data and evaluating the AI methods that claim to find them automatically.
logging weekly
Sociologist turned computational linguist, working where language, data, and culture meet. In the newsroom, I translate journalists’ needs into testable AI systems and evaluation frameworks. In my independent research, I use music charts as records of collective attention. Across both, I turn messy human questions, theories, and data into measurable problems, working tools, and evidence people can understand, without pretending that numbers explain everything.
#01Selected work
A running study of the Spotify UK Top 50 as a record of collective attention, tracking where chart anomalies originate and how cultural moments (a biopic, a World Cup match, an album drop) travel into streaming behavior. Hand-coded theme and mood annotations first, to build gold-standard ground truth before any automated sentiment work.
Long-format weekly data · theme & mood codebook · NLP automation earmarked for v2, evaluated against human labels.
Two early findings: amber marks a moment with an obvious real-world cause:
Building and evaluating language systems and internal AI tools at SWI swissinfo.ch, Switzerland's international multilingual news service. Understanding journalists' needs and issues and translating them into testable AI applications, and evaluating whether the resulting systems are genuinely useful, not merely technically functional.
Journalism · Multilingual · recommendation systems · human-in-the-loop evaluation.
A natural-language hiking search for Switzerland: type “an easy ~10 km loop near Zurich” and an LLM parses it into structured criteria: distance, trail type, surface, loop, which are then matched against the official Swiss hiking network. The interface shows its parsed interpretation of every query, so you always see exactly what the model understood. Working prototype with route stages, maps, and GPX export.
Natural language → structured query · LLM extraction · Swiss open data · Leaflet maps & GPX.
Led the 2024 edition for the Analytics Club: 150+ participants, profit up 130%, and new sponsors signed including UBS and Axpo, after co-leading the 2023 edition across sponsorships, challenge design, and event operations.
Sponsor & stakeholder management · challenge design · event operations at scale.
Co-author of “A Meta-perspective on Reproducibility and Replication Rates” with Heinrich H. Nax: a meta-level look at how reliably empirical findings hold up when re-run.
#02Background
Applied NLP, recommendation, and evaluation of AI systems for a newsroom audience.
Published meta-research on reproducibility and replication rates with H. Nax.
The weekly-chart study above: the analytical, curatorial layer of how music moves through culture.
Grew membership and ran semester events for the student data-science community, alongside leading Datathon (see Selected work).
From social theory to language statistics.
#03Toolkit