I work in data science and analytics — 16 years in retail across personalization, site merch, marketing, customer growth, loyalty, digital ads and experimentation.
Currently at: Walmart
Previously at: TripAdvisor, Chewy, Staples & Macys
Interests: Machine Learning, Agentic AI
📍 San Jose, CA · LinkedIn
A scalable machine learning platform that turns raw customer data (orders, web events, support tickets, marketing touches) into trained, evaluated churn, propensity and uplift models.
Standalone scripts and small apps from marketing and growth analytics problems I've worked on:
| Tool | What it does |
|---|---|
| Marketing Mix Modeling Assistant | A simple OLS + Lasso MMM with budget suggestions — deliberately basic, no adstock or saturation yet |
| LLM Sentiment & Topic Pipeline | LLM-based review classification with topic extraction |
| Breakeven ROAS Calculator | A small Streamlit tool for margin-aware ROAS thresholds |
| A/B Test Sample Size Calculator | Two-sample power analysis for experiment design |
A work-in-progress multi-agent system for subscription churn, designed spec-first: the architecture, agent contracts, and design tradeoffs are documented in the repo, and I'm implementing it phase by phase against that spec. The design covers streaming feature ingestion (Kafka), an XGBoost + LightGBM ensemble with SHAP, Claude-generated explanations, and CUPED-based experimentation.
Python XGBoost Kafka Redis FastAPI Anthropic Claude SHAP CUPED
- The model is rarely the hard problem — feature freshness, auditability, and the decision layer above the model usually matter more.
- Explainability is a product feature. A churn score nobody understands is a churn score nobody acts on.
- Experiments should compound: each outcome should make the next decision smarter.
ML: Python · XGBoost · LightGBM · scikit-learn · SHAP · statsmodels LLM / Agents: Anthropic Claude · OpenAI · RAG Infra: Kafka · Redis · FastAPI · Docker · GitHub Actions Experimentation: CUPED · power analysis · causal lift measurement