This project simulates and evaluates the impact of a new content recommendation engine for Streamly (a fictional streaming platform). It demonstrates a structured causal inference workflow grounded in product analytics.

• Simulated user-level engagement data with realistic behavior across time, cohorts, and segments
• Validated pre-treatment balance using mixed-effects pre-trend models
• Estimated treatment effect via g-formula linear models with covariate adjustment
• Detected a significant engagement lift of ~3.8 minutes per user
• Modeled user-level heterogeneity using random intercepts

Applies core experimentation techniques in a streaming product setting. See the full repository for methods and code.