Product Data Scientist working at the intersection of experimentation, growth, and AI-assisted analytics. I care about turning messy funnel data and ambiguous PM questions into decisions teams can actually act on.
Most of my day-to-day is A/B testing, activation and retention analysis, and building lightweight tooling to make DS work faster and more rigorous.
llm-ds-workflow — A working playbook of prompt patterns I use day-to-day: warehouse SQL drafting, A/B test readouts, exec summaries, null-result framing. Built around Claude + Cursor. These patterns cut my analysis turnaround by ~50% on recurring work types.
activation-insight-agent — A Python agent that ingests funnel data, runs cohort analysis in pandas, and uses the Anthropic API to generate a written drop-off diagnosis. Designed around a specific principle: code does the math, the LLM writes the narrative — because LLMs are unreliable at arithmetic and reliable at prose.
Both are practical tools shaped by real product DS work, not toy demos.
Analytics & data: SQL, Python, Databricks, Amplitude, Avo Methods: A/B testing, causal inference, propensity score matching, forecasting, churn/retention modeling ML: Logistic regression, XGBoost, LightGBM, Random Forest, clustering LLM-assisted analysis: Claude API, Cursor, prompt design, RAG basics