I’m a Causal Marketing Measurement Scientist specializing in Bayesian Marketing Mix Modeling (PyMC), geo-based incrementality testing, and budget optimization.
I build and operationalize scalable MMM frameworks that translate complex posterior outputs into real-world media investment decisions. My work focuses on experiment-anchored modeling, hierarchical geo-level MMMs, and turning decision science into repeatable systems.
- Bayesian hierarchical MMM in PyMC
- Geo holdout experiment design
- Incrementality calibration of MMM
- Budget optimization under uncertainty
- Scalable modeling workflows and automation
Causal & Econometric Modeling
- Bayesian MMM (PyMC)
- Hierarchical modeling
- Bayesian Structural Time Series (CausalImpact)
- Incrementality testing & geo experiments
- Budget optimization & scenario modeling
Programming
- Python (PyMC, ArviZ, Pandas, NumPy)
- SQL
Cloud & Data
- Google BigQuery
- AWS
- Databricks
- MLFlow
Visualization & Communication
- Matplotlib
- Seaborn
- Plotly
- Tableau
Research, Communication, Accountability, Initiative, Collaboration, Critical Thinking, Passion, Presentation, Project Delivery, Idea Generation
- My main portfolio website that contains the following data science projects
- Analyzed the impact of a "Delivery Club" special on sales for a grocery retailer using a Python causal impact library, revealing a 41.1% uplift in sales.
- Developed a predictive model for Parkinson's severity using boosted tree models with feature engineering, resulting in significant improvement in F1 score and recall.
- Analyzed customer buying patterns to uncover product relationships in alcohol retail, revealing insights about customer preferences.
- Predicted customer behavior using historical music sales data, achieving high accuracy through feature space compression and Random Forest.
- Segmented customers for a grocery chain to provide marketing insights based on dietary preferences.
- Created a dashboard for targeted marketing campaigns based on bank customer demographics.
- Boosted tree models for predicting the maximum severity of Parkinsons for clinical patients based on their protein and peptide mass spectrometry quantities.
- Explainable results using Feature Importance for each patient and a description of the top proteins.
- Uses an XGBoost Classifier to predict how likely an employee will leave based on the HR data.
- Provides SHAP values to explain the probability score assigned to the employee.
- Used DenseNet-201 pre-trained neural network and fine-tuned one hidden layer with 4096 nodes and 30% dropout for regularization.
- Solved imbalanced target distribution using class weights to the parameters for the neural network.
- Applied hyperparameter tuning of # of hidden layers, nodes, learning rate, batch size, learning rate decay, and momentum to find optimal values.
- Final Test Statistics:
- AUC: 0.990
- F1: 0.972
- Recall: 0.971
- Precision: 0.973