LLMs & document AI
Extracting and validating structured data from unstructured sources, from historical archives to job ads.
Researcher in economics / data science
Economics,
read through data.
I research how large language models can turn messy, unstructured sources into reliable data for economics. Notes, packages and lecture material from that work live here.
Guide
The modern data-science stack.
A map of how I work, from a fresh laptop to a running pipeline: set the computer up, write code with agents, analyse data and bring it to production. Every stop has an overview of my setup and the posts that go deeper.
Research
Cleaning the whole house.
Presented at the International Conference on Real Estate Statistics 2026, Tokyo
Before a house price index comes the unglamorous part: turning hand-typed, fragmented addresses into reliable locations. I presented how address reconstruction and spatial matching lift the share of linked Budapest transactions from about half to about 93%.
Research
From printed pages to balance sheets.
Presented at the European Historical Economics Society (EHES) 2025
The Magyar Compass is a Hungarian financial yearbook that ran from 1874 to 1944, printed on paper. In the Compass project I extract its financial tables: rules find the balance sheets, a vision-language model reads the hard ones, and every extraction is checked against the totals printed on the page.
Research
Did markets see the war coming?
Published in Finance Research Letters
We used Kyle's model of informed and uninformed investors to ask when the conviction of a war showed up in prices. The answer: markets started pricing it about 50 days before the outbreak, around January 6.
Blog
Ideas, tools & experiments.
Path
Experience & education
Open question
European Central Bank
Data Science Expert, DG/S Financial Market Data
Central Bank of Hungary
Data Scientist / Research Expert
Eötvös Loránd University
PhD in Economics & Finance
Corvinus University of Budapest
MSc Economic Analysis, BSc Applied Economics
About
Trained as a researcher, wired as a developer.
I am a researcher in economics and data science. My current research uses large language models to extract and validate structured information from unstructured sources, from vision-LLM pipelines that digitise historical bank balance sheets to NLP that measures how generative AI exposes occupations. Before that I built large-scale financial data pipelines at the European Central Bank, after several years of empirical research at the Central Bank of Hungary. I am finishing a PhD in Economics & Finance at ELTE. This website shares lecture slides and posts about my research.
Focus
What I spend my time on
Econometrics
Survival analysis, Bayesian models and careful causal thinking applied to markets.
Text & NLP
Measuring sentiment, topics and attention in central bank communication and the media.
Open-source tools
Fast, tested tooling in Python and Rust, plus R packages like currr and ggProfessional.