# Resume Bullets

- Built a sequential causal inference project on KuaiRec to estimate whether high-watch-exposure recommendation days affect future 7-day engagement.
- Defined a user-day estimand, engineered lagged user-history confounders, and diagnosed time-varying confounding and positivity in recommender-system logs.
- Estimated long-term effects using marginal structural models, g-computation, and doubly robust AIPW, with user-cluster bootstrap uncertainty.
- Found small, uncertain average effects on future interaction volume across estimators, while identifying recent engagement and watch-quality histories as heterogeneity drivers for future experiments.
- Produced portfolio-ready causal analysis artifacts, including estimator comparison figures, balance diagnostics, sensitivity tables, limitations, and final product recommendations.