Sylvain Estebe

Sylvain Estebe

Research Assistant · Aarhus University (Center for Humanities Computing)

I build computational models to study how people make decisions under uncertainty — using Bayesian inference, agent-based simulation, and LLM pipelines. My work spans individual cognition, collective decision-making, and cultural evolution, with a common thread: how environmental conditions (resources, volatility) shape behavior.

email bluesky github osf cv

Research

Population · Cultural evolution

Romantic love in French fiction, 1500–1999
LLM ensemble annotation pipeline applied to a corpus of French novels. Tests a behavioral-ecology prediction: does the rise of romantic love in fiction track changes in resource availability? With Nicolas Baumard (Institut Jean Nicod, ENS/PSL).
Manuscript in preparation.

Group · Collective decision-making

Quadratic voting and polarization
Agent-based model of Bayesian cognitive agents comparing plurality vs. quadratic voting under environmental volatility (economic shocks, security crises). Does mechanism design reduce polarization? With Nicolas Legrand (CHC, Aarhus).
Manuscript in preparation.

Individual · Computational cognitive science

A cognitive model of customer churn
MSc thesis. Hierarchical Bayesian model of the decision process behind customer churn — modeling how beliefs about a service evolve and tip into leaving. Used to inform data collection strategy. Industry collaboration with Norlys. Supervisor: Nicolas Legrand.

Other projects

Insecurity and cooperation across 13 countries
Decision-making modeling on public goods game data. Perceived insecurity predicts lower contributions and faster cooperation decline — individuals from high-insecurity nations show lower initial optimism and greater sensitivity to others' choices. Code
Predicting the intention–action gap in childcare
End-to-end ML pipeline (Python) on inequality and childcare data. Full workflow: preprocessing, model fitting, evaluation, visualization. Code
Cognitive distortions in embedding space
NLP pipeline: sentence-transformer embeddings → UMAP dimensionality reduction → HDBSCAN clustering (Bayesian-optimized) on 921 labeled thought examples. Explores whether cognitive distortion categories have a data-driven structure in embedding space. Code & dashboard
Cognitive effort in musical performance
Two internships at the University of Oslo with Laura Bishop and Bruno Laeng. Motion capture, mobile eye-tracking, and EMG to measure cognitive and physical effort in pianists. Multi-modal data analysis with wavelet methods.

Industry

Data science intern at Norlys (Danish energy provider). Customer segmentation from high-dimensional data using clustering and UMAP.

Software

PyHGF
Contributor to the open-source Hierarchical Gaussian Filter library — JAX-based predictive coding for computational psychiatry.

Teaching & outreach

Talks & updates

About

I'm interested in how agents translate information about their environment into behavior — whether the agent is a person, a group, or a population. I like moving between scales, building models that are simple enough to reason about and testable against data.

Before academia I spent a decade as a music producer — composing for Jok'air and Disiz, along with work for ads and theater (certified gold and platinum).

Based in Aarhus, Denmark. Ongoing collaboration at Institut Jean Nicod (ENS/PSL, Paris).

Contact

Open to research collaborations and short-term consulting on Bayesian modeling, LLM pipelines, and behavioral data.

sylvain.estebe@gmail.com · GitHub · ORCID · LinkedIn · Instagram

CV (PDF)

Last updated April 2026 · Built with Astro