Behavioural data scientist. I use rigorous methods from behavioural science, econometrics, and machine learning to make evidence easier to act on—especially where resources are scarce and decisions affect people who rarely get a voice in the room.

I trained as a Behavioural Data Scientist (MSc, Honours; ranked 4th in cohort) at IL3, Universitat de Barcelona. Before specialising, I spent 2.5 years as a Business Analyst at Santander, where I learned how large institutions actually move data from spreadsheets into decisions—and where that pipeline breaks.

Today my work sits at the intersection of international development, social impact, and applied behavioural science. I care about inequality, wellbeing, and transparent measurement: what we claim to know, what the data can actually support, and how to communicate that without overselling causality.

• UN context: A geospatial dashboard for UN ESCWA on Lebanon, built with Leaflet.js and WorldPop-style population thinking, an AI-assisted trend filter, and defensive prompting so outputs stay auditable and safe to share in high-stakes settings.
• Field relevance: Project RYSE—end-to-end analytics in R (clustering, Random Forest, XGBoost, GLM, Lasso, ETL), with strong predictive separation (AUC 0.92) and large effect sizes (Cohen’s d > 1) where the problem demanded it.
• Public-good tooling: A World Happiness Report-style Streamlit dashboard for cross-country wellbeing, drivers, and trends—built for analysts, educators, and anyone who needs linked views without re-running a notebook for every question.
• Behaviour change in the wild: RunCoach AI—FastAPI, PostgreSQL, integrations with Garmin/Strava, and LLM-backed coaching (Groq / Llama 3.3), deployed on Vercel/Render.

Python, R, SQL · FastAPI · PostgreSQL · GIS & Leaflet.js · Streamlit, R Shiny · Tableau, Power BI · Git, Docker · cloud deploy (Vercel, Render)

Roles where data serves people, not vanity metrics: impact evaluation, evidence programmes, behavioural insights teams, and policy-facing analytics in organisations that publish methods, welcome scrutiny, and measure success in lives improved—not slide decks shipped.

Pinned work (in order): lebanon-geo-dashboard · ryse-publico · plataforma-running · -World-Happiness-Report-Dashboard

If you’re hiring for IDinsight, J-PAL, IPA, BIT, Dalberg, a UN agency, or a mission-driven team with similar standards of evidence—I'd like to talk.