Computational social scientist and AI researcher.
I’m a computer scientist by training, have a PhD in computational linguistics and AI, and spent 10 years as a research scientist at Facebook/Meta, building production ML and NLP systems at scale and designing the metrics, surveys and evaluation frameworks used to measure outcomes like polarization, trust and social capital. Most recently, as an Anthropic Fellow, I studied whether experts can keep real oversight of AI agents, or end up going along with them without noticing.
I am driven by a methodological curiosity (my curiosity is not methodological, I am curious about methods). I like finding ways to make sense of complex data.
I’ve worked on topics from international migration to network formation, social capital, and personal naming practices.
I am a co-founder of the data visualization collective Çilek Ağacı.
Writing
Pregnancy Timeline Trends
An interactive, privacy-preserving map of what expecting parents talk about, week by week, from conception to the first months with a baby. Built from 300K+ public forum posts with embeddings, clustering and a cross-encoder reranker.
Does your LLM's star sign make it less predictable? Depends on who built it.
I prompted language models with different star signs to see if their answers got more varied. For Google's models they did, a lot; for OpenAI's they got less varied than with paraphrasing.
Featured
Facebook Movement Range Maps
A privacy-preserving mobility dataset I built at Meta during COVID-19. It became Meta's most-downloaded dataset on the Humanitarian Data Exchange and informed the COVID-19 response of the European Commission, the CDC, California's governor and many others, as well as research worldwide.
Turkish Election Analyses
At Çilek Ağacı, we have analyzed and visualized pretty much all of the Turkish elections and referenda in the last decade and a half.