Loading…
01 / 11
Josh
Fjelstul,
Ph.D.
Data scientist
Machine learning engineer
Quantitative social scientist

02 / 11

Focus.

The hardest part of an ML project isn't the ML.
It's the research design.
I'm a social scientist who engineers ML systems. I help organizations frame business problems, decide on the right research design, and build production-ready systems that solve the problem. I do end-to-end projects — cleaning data, training models, validating models, deploying models and building apps.
03 / 11

Expertise.

End-to-end NLP projects.
I specialize is natural language processing (NLP). I design and build end-to-end, production-ready systems — building data pipelines, designing NLP pipelines, training transformers, deploying models, monitoring models, and building web apps. I have experience building independently and with teams across leading international research institutions.
Every project goes from a carefully-framed question, to a research design, to production code. Working across research design and engineering means nothing gets lost in translation between the two.
04 / 11

Education.

Ph.D.
Political Science (Methdology and International Political Economy), Emory Univerity (2019)
M.A.
Political Science, Emory Univerity (2015)
B.B.A.
Canfield Business Honors Program (CBHP), McCombs School of Business, The University of Texas at Austin (2013)
B.A.
College of Liberal Arts, McCombs School of Business, The University of Texas at Austin (2013)
Certificate
Machine Learning, Stanford University, DeepLearning.AI (2023)
05 / 11

Background.

Academia to industry.
Research to production.
I have 10 years of experience conducting quantitative research, overseeing ML project, and building web apps. I've conducting research at leading universities, including Emory University, Washington University in St. Louis, the University of Geneva, and the University of Oslo.
My academic research was in quantitative methodology and international relations, studying international courts. That meant working with some of the most unforgiving text data in the social sciences: dense treaties and highly-technical court judgments, often across languages. I find that the hardest part of most NLP projects is rarely model architecture. It's understanding the domain, defining and validating concepts for labels, deciding which errors are most problematic, and determining whether a model has actually learned the concept you care about. That's why I always emphasize research design first.
06 / 11

Skills.

TransformersNamed entity recognitionSequence classificationToken classificationSpan classificationSentiment analysisInformation extractionSemantic searchDomain adaptationFine-tuningSentence embeddingsVector databasesFAISSCross-encoder re-rankingTokenizationRandom forestsGradient boostingXGBoostDeep neural networksAttention mechanismsReinforcement learningRecommender systemsClustering (k-means, DBSCAN)Principal component analysisLinear regressionLogistic regressionMultilevel modelsFixed effectsRandom effectsStructural equation modelsTime series analysisBayesian methodsDifference-in-differencesRandomized controlled trialsInstrumental variablesRegression discontinuityMediation analysisConfoundingPotential outcomesCausal identificationD3.jsggplot2FastAPINode APIsDockerReact.jsNext.jsCI/CD pipelinesModel monitoringDrift detectionGoogle CloudVercelPostgreSQLMongoDBREST APIs
07 / 11

Media.

The Washington PostRussian invasion of Ukraine
FiveThirtyEightWorld Cup Database
The TimesWorld Cup Database
DataIsPluralIUROPA CJEU Database, World Cup Database
Agence France-PresseWorld Cup Database
LSE · EUROPPEU political analysis
The MarkupWorld Cup Database
DataCampWorld Cup Database
08 / 11

Showcase.

Loading...
09 / 11

Showcase.

Loading…
10 / 11

Research.

Decision-making in the United Nations General Assembly: A comprehensive database of resolution-related decisions

The Review of International Organizations

Joshua C. Fjelstul, Simon Hug, Christopher Kilby2025

The timely administration of justice: using computational simulations to evaluate institutional reforms at the CJEU

Journal of European Public Policy

Joshua C. Fjelstul, Matthew Gabel, Clifford J. Carrubba2023

The CJEU Database Platform: Decisions and Decision-Makers

Journal of European Public Policy

Stein Arne Brekke, Joshua C. Fjelstul, Silje Synnøve Lyder Hermansen, Daniel Naurin2023

The CJEU Database Platform: Decisions and Decision-Makers

Journal of Theoretical Politics

Joshua C. Fjelstul2023

Explaining Public Opinion on the Enforcement of the Stability and Growth Pact during the European Sovereign Debt Crisis

European Union Politics

Joshua C. Fjelstul2022

How the Chamber System at the CJEU Undermines the Consistency of the Court's Application of EU Law

European Union Politics

Joshua C. Fjelstul2021

Improving the efficiency of pretrial bargaining in disputes over noncompliance with international law: encouraging evidence from the European Union

Journal of European Public Policy

Sivaram Cheruvu, Joshua C. Fjelstul2021

The Politics of International Oversight: Strategic Monitoring and Legal Compliance in the European Union

American Political Science Review

Joshua C. Fjelstul, Clifford J. Carrubba2018
11 / 11

Work.

Open to full-time roles in NLP, ML engineering, and applied data science. I bring something most ML teams don't have: rigorous methodological training to design what gets built, not just build it. Based in Austin, Texas. Open to remote.

Scroll