Eleven shipped analyses across operations, quality, and BI — each built to answer a business question, with the source and a live artifact one click away.
An end-to-end dbt + Snowflake pipeline transforming raw Olist e-commerce CSVs into a tested star schema.
Weather structure alone predicts grid load — a feature audit exposed that a single autoregressive lag was inflating the baseline by ~65%, so the champion was rebuilt without it.
Does qualification fit actually predict interviews? An instrument that scores postings on two independent axes and logs real outcomes — with a guardrail that refuses to coach embellishment.
With a finite retention budget, who do you fund first? Ranking roles by cost and by exit rate produces two different lists — Lab Techs are the only role that tops both.
One improvement team, one cycle — where do you point it? Screened sixteen candidate factors, rejected twelve as noise, and isolated three controllable drivers.
A multi-page executive suite linking fulfillment to profit, with computation pushed upstream to a BigQuery data lake for zero-lag cross-filtering over millions of rows.
Loaded 32.4M Instacart order-product rows into a local PostgreSQL warehouse, then mined reorder timing, product co-occurrence, and purchase-interval signals — with every proxy labeled as one.
What drives persistent pandemic burden? A county-level mortality analysis identifying structural conditions — not age or race — as the primary predictors.
Three public datasets, one clean analytical base table — schema mismatches, type conflicts, and FIPS-key formatting all resolved in production ETL.
Length of stay, billing, and predictive modeling across 55,500 records — including a null model result documented honestly rather than quietly dropped.
Fitbit wellness patterns translated into product strategy — weekend activity peaks point to a weekday-nudge opportunity, backed by simple correlation evidence.