Projects
Global Poverty Meta-Analysis Field Extraction
Reduced scholarly-article coding costs from $1,000/paper to $0.12/paper by building an AI-native extraction pipeline (GROBID + LLM) with the Global Poverty Research Lab (NW Kellogg), processing 213+ development economics RCTs.
YouBook
Built an AI-native study platform (Granola-style CRM for students) that auto-generates textbook-quality lecture notes. Placed top five at the Continual Learning Hackathon hosted by Intercom.
Comparing Confidence in LLMs
Produced a reusable benchmark for follow-on calibration research by measuring how LLMs revise confidence estimates as task difficulty scales. Built with PyTorch and HuggingFace.