Browse my

Projects

Churn Prediction

User Churn Prediction Model

End-to-end churn prediction using XGBoost and SHAP to identify at-risk fintech users — drove a 22% reduction in 90-day churn and ~₹8Cr in retained revenue.

Python XGBoost SHAP SQL Optuna
Marketing Mix Model

Bayesian Marketing Mix Model

Built a Bayesian MMM with adstock and saturation to measure true marketing ROI across 6 channels — improved blended efficiency by 31% and freed ₹3Cr in reallocated spend.

Python PyMC Bayesian statsmodels
Uplift Modeling

Uplift Modeling for Incremental Targeting

T-Learner uplift model that identified the truly persuadable users — not those likely to convert anyway. Shifted ₹4Cr of campaign spend off "Sure Things" and onto Persuadables, lifting incremental conversions by 19%.

Python XGBoost EconML Causal Inference Qini Curve
Multi-Touch Attribution

Multi-Touch Attribution Engine

Markov-chain attribution across 7 channels on 2.5M user journeys — exposed that display was undercredited 2.5× and paid search overcredited 1.6×. Reallocated ~25% of paid budget; same spend, 18% lift in attributed conversions.

Python Markov Chains Shapley Values BigQuery NetworkX
Revenue Simulation

Revenue Simulation & Forecasting

Monte Carlo simulation engine that models 36-month revenue under uncertainty — surfaced churn as 2× the lever of acquisition, reshaping ₹30Cr+ annual planning.

Python Monte Carlo NumPy Plotly
Anomaly Detection

Payment Metrics Anomaly Detection

Real-time anomaly detection system for payment success rates, transaction volumes, and latency — reduced mean time to detect critical payment issues by 70%.

Python Isolation Forest statsmodels BigQuery Pub/Sub