Posts by Collection

Soccer Player Market Value Scout

Deployed ML service that predicts soccer players’ market value from performance stats and explains every prediction. Streamlit app + FastAPI on Google Cloud Run.
STAT 418 final project.

mHTI Discussion Analysis Pipeline (M.S. Thesis)

LLM multi-agent simulations of real research teams with a four-layer fidelity framework and a 55-metric NLP evaluation suite. Basis of an EMNLP 2026 submission.

Congressional Stock-Trading Analysis

Analysis of U.S. House members’ reported stock trades, joined to party affiliation, to study trading patterns across parties.

Enhancing Emotion Recognition in AI

Improving the interpretability and reducing the bias of an image emotion-recognition model using CLIP-Dissect and concept-based analysis.

Multimodal LLM Merging (Samsung Electronics)

Layer-wise merging strategies and a scalable safetensors pipeline that raise language-benchmark scores while preserving multimodal performance.
AI Model Research Internship.

Revealing Team Project Development via Conversation Analysis

Published in Working paper, 2025

Analyzing multi-session interdisciplinary team discussions to understand how project ideas evolve from exploration to outcomes, modeling insider–outsider roles, conversational centrality, and turn-taking dynamics.

Recommended citation: Sungjin Choi. (2025). "Revealing Team Project Development via Conversation Analysis." Working paper.

LLM Multi-Agent Simulations of Research Teams

Published in Under review, EMNLP 2026, 2026

LLM multi-agent simulations of five real research teams (24K utterances, 6 sessions/team) with a four-layer fidelity framework. Behaviorally grounded personas close 64% of the naive-to-human outcome gap across 5/5 teams.

Recommended citation: Sungjin Choi. (2026). "LLM Multi-Agent Simulations of Research Teams." Under review, EMNLP 2026.