I am a computational social scientist. Most of my work comes down to one stubborn problem: the things that matter most in politics — attention, affect, identity, trust — are the hardest to measure. I build and test instruments for measuring them. Lately the instrument I am most interested in is the language model itself: I evaluate what it does from the outside, and open it up to see how it works on the inside.
From economics to computational social science
I trained as an economist, and statistics was my first language: models, estimation, causal inference. But the questions that held me — identity, ideology, how political communities form and fracture — kept slipping through the categories that language could express. So I went the other way, into qualitative inquiry: reading texts closely, taking interpretation seriously, learning that what a document means is not something you can tabulate.
Each tradition exposed the other’s limit, and the limits pointed somewhere. Close reading does not scale; counting does not interpret. Computational text analysis was the bridge — treating text as data without abandoning the interpretive questions. And once texts became data, a further step suggested itself: texts are produced by people in relation to one another, and the relations carry as much politics as the words. That is what pulled me into network analysis — the idea that structure itself is measurable, and that it reveals what individual attitudes cannot.
The threads fused in 2018, at the Summer Institute in Computational Social Science: the point where a set of borrowed methods became a discipline I could stand in, and where I stopped only using instruments and started building them. The most recent turn follows the same logic one step further. Machine learning entered my work as an instrument — annotation and measurement at scale — and then became a subject: interpretability, asking of the model what I had learned to ask of texts and networks. What does it attend to? What does it encode? Does it measure what we think it measures?
Each layer kept the ones beneath it: statistics under everything, interpretation disciplining measurement, texts and networks as the substance, models now both tool and object. (A more personal telling of this story is on Substack, in Turkish: Hesaplamalı Sosyal Bilimlere nasıl başladım?)
Before the PhD I worked in the space between research and policy: as a research assistant at a non-partisan foreign-policy think tank in Ankara, and later building the Turkish Foreign Policy Barometer on Twitter, a German Marshall Fund–funded project tracking elite discourse. Those years taught me that measurement is never just technical — what you count shapes what governments and publics believe is happening.
What I do now
At Northeastern, I am a PhD candidate in Political Science with a minor in Computational Social Science, working with Nick Beauchamp and David Lazer. My dissertation, Attention Segregation, asks what organizes the segregation of citizens’ private information worlds — not whether people ever meet the other side, but the character of what they actually attend to, and how far beyond politics the division reaches.
Around the dissertation, the research runs along two lines. The fragmentation work the dissertation anchors traces where political attention, affect, and identity pull apart — from anti-Americanism in Turkish social media to transnational expert networks to legislative speech on technology. The interpretability and evaluation work turns the lens on language models — what their annotations get wrong about political constructs, and where ideology lives in their internals — grounded in graduate training in mechanistic interpretability with David Bau.
Building alongside the research
I have spent years doing two things at once: the research, and building the people and programs around it. I attended the Summer Institute in Computational Social Science in Helsinki as a fellow in 2018, came home, and co-founded SICSS Istanbul, co-organizing it from 2019 to 2023 as one of just two organizers and training cohorts of early-career social scientists. I co-founded BLISS — the Boston LLMs Initiative for Social Sciences — a funded workshop series bringing LLM methods to social scientists. I co-designed and led the AIDE summer bootcamp at Northeastern’s Ethics Institute, mentoring fifteen philosophy and computer-science graduate students one on one, from their first line of Python to building with LLMs. And at the Boston Area Research Initiative, I consulted for civic organizations on the City of Boston’s open data.
The two halves are the same commitment at different scales: the research measures how communities of knowledge form; the building grows them.
Off the page
I grew up in Turkey and have lived between Istanbul, Ankara, and Boston. I speak Turkish and English, and get by in Arabic. I write occasionally at The Social Computist, post short updates on LinkedIn, and keep a running digest on the Now page.
Curriculum Vitae
The full record — education, experience, publications, talks, awards, and references — is in the CV (PDF).
- PhD, Political Science (minor: Computational Social Science) — Northeastern University, expected Spring 2027. Committee: Nick Beauchamp (advisor), David Lazer, Risa Kitagawa.
- PhD coursework, International Relations — Kadir Has University, 2019–2021
- MSc, Middle East Studies — Middle East Technical University, 2019
- BA, Economics — Bogazici University, 2016