Confusion Bank
An AI study companion that identifies and resurfaces a learner's confusion points from historical conversation data. Built as a solo project for HackMIT 2025.
Hi, I’m David.
I’m a senior at Haverford College studying Computer Science and Mathematics (Statistics focus).
I study how intelligent systems can learn from both collective and individual experience to better understand and empower humans. My research vision — “Experience as Intelligence” — explores how AI integrates structured, population-level data with personal, interaction-level experience to adapt across time and context. I’m broadly interested in how this principle enables AI to generalize to the population while personalizing to the individual.
I’ve applied this perspective to domains such as human-centered AI, including work in education and AI ethics/safety. Methodologically, I draw on statistical modeling, deep learning, and reinforcement learning to connect population-level reasoning with adaptive, human-centered intelligence.
Research
LLM Moderation Auditing and Steering — Haverford College
- Developing large-scale auditing pipelines to study how language models shift moderation behavior over time.
- Built genAIaudits.github.io for longitudinal monitoring of acceptable speech boundaries.
- Co-author on Longitudinal Monitoring of LLM Content Moderation of Social Issues (CHI 2026 submission in preparation).
- Senior thesis: designing interpretable concept-based steering methods to adjust model moderation stance while preserving semantic fidelity.
Human-Centered AI in Education — University of Pennsylvania
- Analyzed learner behavior across 200+ MOOCs to identify course-level determinants of dropout.
- First author on Understanding MOOC Stopout Patterns: Course and Assessment-Level Insights, published at Learning@Scale 2025.
- Evaluating LLM-based teaching assistants (JeepyTA) in live instructional deployments to understand feedback effects on learning.
- Comparing Bayesian and deep knowledge tracing approaches to model evolving student mastery with interpretable mechanisms.
Automatic Grading Methods — Bryn Mawr College
- Reproduced and extended generative grading models to small transformer-based architectures for multimodal assessment.
- Built a probabilistic data simulator to support evaluation across both text and visual student responses.
- Source Code
Selected Projects
PA Solar Energy Snapshot
Led a data engineering project for the Philadelphia Solar Energy Association to automate data ingestion and deploy public-facing interactive dashboards.
Intelligent Tutoring System
Designed and implemented an intelligent tutor for JavaScript using the CTAT framework as part of the Carnegie Mellon University LearnLab Summer School.
Activities & Leadership
- Gold Prize, Tencent AI Arena Global Open (2025): Led a team to develop a PPO-GRU agent for a POMDP environment, placing in the top 1.6% of 1000+ undergraduate teams.
- Top Prize, DataFest Philly (2025): Led a team in analyzing commercial leasing trends post-pandemic, awarded for innovative use of external data sources.
- Data Analyst Intern, Roundhouse One (2024): Modeled HVAC effects on indoor environmental quality across 200+ K–12 schools in Hawaii using statistical and causal inference methods.
- Senior Thesis in Mathematics: Developing theoretical results on the Poisson binomial distribution and its extensions to sequentially dependent trials.