Projects

This thesis investigates how large language models (LLMs) behave in social interactions through a systematically designed Trust Game experiment. The study simulates decision-making by pairing LLM agents with human behavioral models, examining the LLM’s cooperative tendencies across different personas (nationality, gender), models (OpenAI, DeepSeek, Mistral), and prompt languages (English, Chinese, French). Controlled variables such as temperature, game structure, and role assignment ensure robustness and internal validity.

This project develops an automated data classification and integration pipeline for the Future Time Orientation Life Goal project, leveraging large language models (LLMs) to categorize individuals’ life goals into structured taxonomies. I contributed to data annotation, prompt engineering, and codebook refinement, improving cross-language consistency across multiple-country datasets. Additionally, I implemented Python scripts for data merging, inter-coder reliability analysis, and anomaly detection. The project also explored Web-App prototyping using Next.js and Supabase, building an early-stage interactive platform for model-based classification and data management.