Curriculum vitae

Zhongyi Tang

Research Data Scientist · Google

zhongyi418@gmail.com · zhongyitang.github.io

Applied scientist with a background in economics and machine learning. My work spans AI agents for analytics, LLM evaluation, experiment design, and causal measurement.

Experience

Present

Research Data Scientist

Google

2024–2026

Economist II

Amazon

Built AI agents for analytics and experimentation, alongside methods and infrastructure for measuring marketing impact.

  • Developed agents for natural-language experimentation workflows, text-to-SQL, and automated analysis and reporting, with persistent memory and a separate LLM-as-a-judge evaluator.
  • Designed and analyzed regional holdout experiments using synthetic difference-in-differences; improved precision and calibrated observational models against experimental results.
  • Parallelized experimentation pipelines to reduce runtime by a factor of eight and expanded capacity from 3 to 12 marketplaces.
2017–2019

Research Analyst

Stanford Institute for Economic Policy Research

Built software pipelines to export satellite imagery at scale, reducing processing time fivefold. Trained and calibrated convolutional neural networks using satellite imagery and household survey data to predict infrastructure outcomes in Africa.

Education

Ph.D. in Economics

Boston University

B.S. in Economics (Honors) & Mathematics

Trinity College

Methods & tools

Research
AI agents, LLM evaluation, causal inference, econometrics, experimentation, machine learning
Programming
Python, R, SQL, PyTorch, TensorFlow, scikit-learn, pandas, NumPy
Data & systems
AWS (S3, EMR, EC2, Lambda, Step Functions, Athena), PySpark, Hive, Hadoop

Publications

AEA Papers and Proceedings 2026

A New Perspective on Spatial Heterogeneity in African Development

Pascaline Dupas, Nicolas Suarez, and Zhongyi Tang.

All authors contributed equally.

We combine satellite imagery and survey data to map infrastructure access across Africa using machine learning. We then use spatial regression discontinuity to estimate the impact of national institutions on infrastructure access.

Read paper Data & code

Nature Communications 2020

Using publicly available satellite imagery and deep learning to understand economic well-being in Africa

Christopher Yeh, Anthony Perez, Anne Driscoll, George Azzari, Zhongyi Tang, David Lobell, Stefano Ermon, and Marshall Burke.

We combine deep learning with public daytime and nighttime satellite imagery to predict local asset wealth across sub-Saharan Africa. Trained on survey-based wealth measures, the models are evaluated on unseen locations and on changes in wealth over time.

Read paper