Vansh Gupta 42.3608 N, 71.0843 W · Cambridge, MA

⊕  Software

Tools built en route

Most of these exist because a project needed data or a method I couldn't find anywhere else. The code is on GitHub.

01Tools

STE: Strategic Treatment Effects

R package · random forests and LASSO

The same business strategy can help one company and hurt another. STE estimates who benefits from a choice and who is harmed by it, using observational data with many variables. It goes with Jorge Guzman's paper Treatment Effects in Managerial Strategies. Random forests estimate how likely each firm was to make the choice, local regressions estimate the effects, and LASSO picks out what drives the differences.

Upload-to-Zenodo

Python · used by the AEA Data Editor's replication team

Keeps research files in sync between a computer and Zenodo, the public research archive. I wrote it while checking papers for the American Economic Association, and their replication team of about 40 people still uses it.

Gazette reader

language-model pipeline · 16.4 million paragraphs so far

The measurement system behind the Brazil state-capacity project. Cheaper language models sort the paragraphs first, stronger models handle the hard cases, and a separate labeled set is used to check accuracy. The code will be released with the papers.

Back to research →