Research
Systems
Language models for messy evidence
Language models for messy evidence
I build research workflows that turn survey responses, conflict reports, and county-scale data into inspectable evidence. The work sits between machine learning and domain research: multi-agent thematic analysis, domain-adapted NLP, multilingual conflict analysis, and spatial modeling.
Role
AI/ML Researcher
Research Surface
Text, conflict data, counties
Stack
Python, LangGraph, Transformers
Output
Systems, analysis, papers
The research stack includes a LangGraph pipeline that reduced expert analysis time by 10x, BERT fine-tuning that reached 90 F1 on conflict-text analysis tasks, Spanish and Arabic extensions for cross-lingual analysis, and a county-disparities study across 34 outcomes, 78 predictors, and 3100+ counties. Two papers from this work are under review.
Study Area
Applied AI Research
Publication Status
Two papers under review
Primary Focus
Research infrastructure
Program Fit
Interdisciplinary systems