About Me
Researcher and product data scientist with a decade spanning user behavior experimentation at Amazon, causal inference research at Columbia, and product analytics and engineering for startups.
Research
Columbia Business School
Running an online experiment (oTree, Heroku, Prolific) to measure how developers make AI-code adoption decisions in the IDE, and a difference-in-differences causal analysis of how AI is changing experimentation practices across 2,000+ firms. Other projects include an RCT meta-analysis and data engineering for analyzing EV charging behavior nationwide.
Data Extraction and Causal Analysis of Cloud Seeding
First author of "Structured dataset of reported cloud seeding activities in the United States (2000–2025) using an LLM", published in Scientific Data (Nature Portfolio) (link). Built a PDF extraction pipeline with OpenAI integration to process 800+ scanned NOAA cloud seeding reports into a structured dataset, helping address the gap in cloud seeding data. Now conducting a within-site difference-in-differences analysis to estimate the causal effect of historical cloud seeding operations on precipitation: Difference-in-Differences Explorer. Presented the methods behind this work in a guest lecture at the CUNY Graduate Center for 15 PhD and master's students, covering practical workflows for using large language models in scientific research: Guest Lecture: Using LLMs for Science (PDF).
Generative AI Can Harm Learning: A Reproduction and Extension
Reproduced the three-arm cluster-randomized trial from Bastani et al. (2025, PNAS) on GPT-4's effect on high school math learning, simulating 943 students across 50 classrooms in R to validate the causal identification strategy. Extended the analysis on author-shared data with covariate balance diagnostics, classroom-clustered standard errors, and heterogeneous treatment effects by prior AI exposure. GitHub, Quarto Report, Presentation (PDF).
Behavioral Economics Research Proposal: Strategic Route Choice in Navigation Apps
Behavioral economics research proposal testing for Level-K thinking in how drivers choose routes when navigation apps recommend paths that depend on other users' behavior: PDF.
Product Data Science and Engineering
Amazon Alexa+ Smart Display Experiences
Improved device engagement and customer satisfaction with Alexa's smart display devices. Launched and measurably improved the quality of the new multimodal screen experiences for Alexa+.
Amazon AlexaAnswers.com
Led 8x growth in monthly active users through product experiments and marketing expansion. Read more here.
Product Science, LLC
Founded Product Science Consulting to help early-stage startups get from insights through measurable impact. Partnered with the founder of an interior design technology company to improve product strategy, setup analytics infrastructure end to end, and deploy customer-facing AI features. Running DesignMyExperiment.com to provide free help with experiment design, including curated libraries of real product A/B tests and methodology reviews.
Education
