Experience

Google Public Policy Fellow

Evaluating whether frontier models track how national cultures actually change.

• Led an evaluation of how frontier LLMs represent national cultural change over time.
• Benchmarked Claude, GPT, Gemini, and Qwen against two decades of World Values Survey data across 40 countries.
• Designed the evaluation framework, including construct operationalization, prompting protocols, and statistical analysis.
• Contributed research to a report on multilingual and multicultural safety risks of LLMs.
• Contributed to a CDT blog post on the social and cultural risks of sovereign AI.

Outputs: arXiv preprint (first author) · multilingual safety report · CDT blog post

Trustworthy AI Researcher

How AI systems can be independently assessed and certified.

• Researched how AI systems can be independently assessed and certified as trustworthy.
• Conducted comparative analysis of international AI governance frameworks.
• Co-authored a report on independent certification for trustworthy AI.

Outputs: Report on independent certification for trustworthy AI

Graduate Researcher

Modeling how moral and cultural values change, and how AI use changes people.

• Building agent-based models of moral and cultural value change.
• Studying how interacting with AI shapes identity and sense of self.

Graduate Researcher

Computational and experimental research on morality, cultural difference, and how AI systems represent both.

• Studied how LLMs influence cultural values, communication norms, and knowledge systems, tracing homogenization in model outputs to training-data imbalance and culturally narrow evaluation pipelines.
• Wrote guidelines for developers, policymakers, and research institutions on building culturally representative, safety-aligned systems.
• Built NLP pipelines for over 10M+ social media posts with engineering collaborators, using fine-tuned transformer models to classify stance and moral framing.
• Designed annotation systems end to end, wrote coding guidelines, combined human coding with AI-assisted labeling, and led native-speaker teams for a multilingual moral reasoning benchmark in English, Persian, Italian, and Portuguese, quantifying cross-lingual misalignment in LLM moral reasoning.
• Designed and translated multi-language surveys for a DoD-supported study of cultural norms and social evaluation, establishing conceptual and measurement equivalence across populations.
• Ran preregistered experiments (N > 2,300) on how partisans misjudge each other's moral views, then tested a feedback intervention that reduced misperception and increased intergroup trust, using regression, ensemble classifiers, and MANCOVA.

Outputs: “The Homogenizing Engine” (2025) · MFTCXplain, EMNLP Findings 2025 · misperceptions manuscript under review

Researcher

Globally distributed replication and multi-site studies.

• Contributed to large-scale multi-site studies on social judgment, risk perception, and decision-making, run by a network of 2,400+ researchers across 70+ countries.
• Built automation for survey deployment and reporting, cutting study launch time by roughly 70% and reporting turnaround by roughly 50%.
• Recruited and analyzed data from 1M+ participants, applying rigorous quality checks, mixed-methods analysis, and cross-cultural validation techniques to ensure reliable insights.
• Synthesized results for international teams and recommended methodological changes that fed into study redesigns.

Outputs: Global Patterns in Moral Judgment · Semantic priming across many languages

Research Assistant

Experimental research on bias in social judgment.

• Contributed to a large-scale behavioral study on reducing attractiveness-based bias in social judgment and decision-making.
• Reviewed the literature and helped design an evidence-based intervention, building study materials and experimental tasks.
• Assisted with data interpretation and manuscript preparation on the study's implications for person perception and social evaluation.