James Tavita

• Founded and building evaluation infrastructure for AI agents and the people who work alongside them.
• Own product direction, system architecture, and evaluation methodology end to end.

Background in Economics and Computer Science

Identifying Treatment Effects Under Selection

An interactive walkthrough of a full causal identification workflow — DAG, identification argument, difference-in-differences estimator, and robustness battery — built on clearly labeled synthetic data. (Illustrative demo built on synthetic data.)

Scott Condie, Gabriel Lehnardt, James Tavita (2026). Oversight Risk: How Committees Shape Portfolios. Journal of Portfolio Management.

Tom Hunsaker, Abdulaziz Alakeel, James Tavita (2026). From Payments to Power: How the PayPal Mafia Shaped Silicon Valley's Venture Landscape. Thunderbird Case Series, Harvard Business Publishing (forthcoming).

Censiq — Evaluation Infrastructure for AI Agents

Founded and built Censiq to give organizations a defensible, evidence-based way to evaluate AI agents before and after deployment.

Identifying Treatment Effects Under Selection — An Illustrative Study

Built a full interactive walkthrough of a causal identification workflow — DAG, identification argument, estimator, and robustness battery — on clearly labeled synthetic data.

Model Comparison & Causal-vs-Predictive Playground

Built an interactive playground comparing model families on synthetic data and demonstrating why the variable most useful for prediction is not always the variable that should be interpreted causally.

NBA Player Performance Forecasting: CatBoost vs. Random Forest

Built and backtested two full-season forecasting models against real 2024–25 results; CatBoost averaged R² = 0.54 vs. Random Forest's 0.44, and ranking players by how far they beat the model's forecast placed the actual Most Improved Player winner at #2 of 218 eligible players.

Who Leads a Mission — A Study of 504 Mission Presidents

Built a reproducible pipeline and an interactive write-up showing that supply is statistically proportional to membership across US states but strongly disproportionate between countries, and that international leaders are ~4.5x more likely to have previously led a stake after adjusting for age and family size.