This project simulates an A/B test for Streamly (a fictional streaming platform) to estimate the engagement lift from a new recommender system. It applies core product data science methods—pre-treatment balance checks, covariate-adjusted models, and sensitivity analyses—to demonstrate how causal inference guides feature rollouts in production.

• Simulated user behavior across time, cohorts, and demographics
• Applied generalized estimating equations (GEE) for repeated measures causal analysis
• Controlled for pre-treatment behavior via merged historical engagement
• Detected statistically significant lift in engagement post-treatment
• Includes modeling, diagnostics, and feature importance analysis (via XGBoost)

For full code and methodology, see the GitHub repository.