Bradley DePanfilis BDepanfilis

Bradley DePanfilis — Data Analytics & Data Science

Developed a reproducible ETL pipeline using BigQuery SQL to compute two consumer price indices (fixed-weight Laspeyres and chained Laspeyres) tailored to Gen-Z spending, validated against BLS CPI-U.
Deliverables: versioned CSVs, GitHub Pages site with Plotly visualizations, and a Power BI dashboard.
Highlights: Transparent methodology, normalized base (2020=1.00), cost-aware query design, and automated data processing for easy re-runs.
Skills: SQL (BigQuery), data pipeline development, data visualization, time-series analysis, index construction.
Repo: REPO_URL • Live: LIVE_URL • Power BI: LIVE_URL

Analyzed CPS March ASEC microdata to estimate the U.S. career gender wage gap using sequential regression models (baseline, education, personal/geo, household, singles).
Implemented HC1 robust standard errors, year-by-year trends, and publication-quality visualizations with R/ggplot2.
Deliverables: Modular R scripts (wrangle/analysis), CSV outputs, GitHub Pages site, and Power BI dashboard.
Highlights: Reproducible data wrangling, robust statistical modeling, and clear visualizations showing a persistent 21–24% gap (14% for singles).
Skills: Statistical modeling, R (tidyverse, broom, sandwich/lmtest), data visualization, data wrangling, regression analysis.
Repo: REPO_URL • Live: LIVE_URL • Power BI: LIVE_URL

Built an interactive e-commerce analytics dashboard using Streamlit, DuckDB, Altair, and Pandas on the Olist dataset, delivering metrics on revenue, AOV, customer cohorts, LTV, retention heatmaps, returns, and marketing ROI (ROAS).
Highlights: Optimized query caching, defensive data handling, and production-ready secrets management; serves ~180 days of data with snappy performance.
Skills: Analytics engineering, Python (Pandas, Altair), Streamlit, DuckDB, data visualization, ETL pipelines, cohort analysis, LTV modeling.
Repo: REPO_URL • Live: LIVE_URL

Engineered a curated analytics warehouse using dbt Core to transform raw Olist e-commerce data into a star schema with staging, core dimensions/facts, and marts for sales, marketing ROI, cohorts, and returns quality.
Implemented data quality tests (dbt_utils, dbt_expectations), seeds for enrichment, and automated dbt Docs publishing to GitHub Pages for lineage and observability.
Highlights: Warehouse-agnostic workflow with DuckDB for local/CI and optional BigQuery execution with cost guards; demonstrates robust ELT and CI/CD integration.
Skills: Data engineering, dbt Core, SQL, DuckDB, BigQuery, data modeling, ELT pipelines, data quality testing, CI/CD.
Repo: REPO_URL • Live Docs: LIVE_URL

Created an end-to-end A/B testing pipeline for uplift modeling using the Criteo Uplift dataset, featuring power checks, causal effect estimates (diff-in-proportions, CUPAC, IPW/AIPW), and targeting via Qini/AUUC with a Top-K policy.
Engineered as a reproducible Python package with CLI entry point, deterministic seeds, and auto-generated reports/figures; includes covariate balance (SMDs) and propensity diagnostics.
Highlights: Narrative Jupyter Notebook with ROI appendix; supports data-driven marketing and product decisions.
Skills: Data analytics, Python (pandas, NumPy, statsmodels, scikit-learn, matplotlib), causal inference, A/B testing, statistical modeling, data visualization.
Repo: REPO_URL

Developed a production-ready Bayesian team rating system in for Valorant teams displayed on Vlr.gg, aggregating player stats into team skills with hierarchical priors for uncertainty, incorporating margin-of-win scaling, map adjustments, meta updates, roster resets, time decay, and regional biases to enable accurate 1:1 cross-regional predictions for Tier 1 & 2 teams across global regions. .
Deployed for 2024–2025 VCT/VCL matches with >28M total visitors; supports dynamic leaderboards and real-time updates.
Highlights: Modular design with FastAPI, SQLAlchemy, and PostgreSQL; deployed via Docker on AWS; integrates Chart.js for visualizations.
Skills: Data engineering, Docker, AWS, Python (pandas, NumPy, FastAPI, SQLAlchemy, Pydantic), PostgreSQL, API development, Bayesian modeling, data visualization.
Repo: REPO_URL

Built a lightweight GitHub Pages app using JavaScript and a simple Elo system to rank cake preferences, demonstrating decision-making tools with a playful dataset.
Highlights: Clean UI, reproducible logic, and rapid deployment for stakeholder interaction.
Skills: Data analytics, JavaScript, data visualization, decision support systems.
Repo: REPO_URL • Live: LIVE_URL

Repos are structured for quick digestion. If you’d like a private walkthrough of any project or a short Loom overview, I’m happy to share.