Software and Data
I develop software and datasets for economic research and policy analysis. The projects below include AI assistants, research software, and open data. Each entry describes my contribution and links to the project.
I created Ask Atlas to help people query and interpret international trade data from the Atlas of Economic Complexity in plain language. I built, deployed, and maintain the agent, web application, and evaluation framework. The assistant identifies the countries and products in a question, retrieves relevant methodology, writes and runs SQL queries, and links answers to Atlas visualizations. Answers are checked through verification steps and a curated set of evaluation cases.
Ask Atlas is available through an early-access sign-in process.
I built Growth Lab Deep Search to provide an AI-assisted knowledge management system for the Growth Lab. The tool helps researchers ask natural-language questions over the Growth Lab’s research outputs. It finds and ingests Growth Lab research outputs, tags these papers using AI, and supports retrieval, answer generation, access controls, evaluation, and citation-verification tools.
Researchers can use the website or connect an AI client through the Model Context Protocol (MCP). Access requires sign-in and approval, and some source material has additional restrictions.
I developed the Complexity AI Tutor as an AI-assisted learning platform for economic complexity. I piloted it at the July 2026 Economic Complexity Summer School at UM6P Morocco, and it now supports self-study. Learners can work through concepts with Socratic coaching, and review material through AI-assisted quizzes and spaced repetition.
The platform includes explanations and analytics that help instructors review usage and misconceptions. Access requires sign-in and approval.
I wrote py-ecomplexity, a Python implementation of the Hidalgo and Hausmann economic-complexity methodology. Researchers can use it with trade, employment, and other activity data to measure specialization, complexity, and relatedness, and to identify nearby diversification opportunities. The package is used in the Atlas of Economic Complexity and by research groups in many countries, and it has been downloaded more than 37,000 times from PyPI as of November 2026.
With José Morales-Arilla, I developed GLocal, a dataset for comparing economic and environmental conditions across subnational regions worldwide. I built the reproducible pipeline that combines satellite and geographic data, including nighttime lights, population, climate, and infrastructure indicators.
The Scientific Data article describes the dataset and its validation. Data and code are available through Harvard Dataverse.
I developed econ-embeddings to provide a repository of carefully generated embeddings for codes in economic classifications, including products, technologies, industries, and scientific fields. The project uses a language model to write fuller descriptions of each code, including disambiguation from neighboring codes, then converts those descriptions into semantic embeddings that support searches and matches across classifications. The repository provides the embeddings and the pipeline for adding other classifications.