Welcome to my data science lab! This repository hosts my data science case studies, automated data pipelines, and interactive dashboards.
See DEVELOPMENT.md for the pinned Python, uv, and Quarto environment and isolated validation command.
This site is architected to separate deep-dive engineering work from visual analytics:
projects/(Case Studies): Full-stack data engineering projects with Python pipelines, automated testing, and written analysis.dashboards/(Visualizations): Hosted Tableau/PowerBI embeds and interactive Plotly apps.projects/oregon-fqhc/tests/: Automatedpytestsuites ensuring data integrity before deployment.
A reproducible annual snapshot of Oregon's HRSA-reported health-center landscape. It combines a dated site-footprint extract with organization-level 2024 UDS measures, identifier-based joins, validation, and transparent analytical reporting.
This project demonstrates a scheduled ETL pipeline with explicit validation gates running entirely on GitHub Actions:
- Ingest: Python scripts retrieve authoritative, separately dated source snapshots from the HRSA Data Warehouse and FOIA reading rooms (2024 UDS patient measures). These are not live EHR counts.
- Transform:
pandasperforms cleaning, geospatial field standardization, and validated many-to-one joins. Current coverage and precise counting definitions are documented in the project data contract. - Validate:
pytestseparates hard integrity failures from visible reconciliation warnings and analytical-review triggers. - Analyze and publish: One equally weighted row per organization supports an exploratory log-scale size/Medicaid-share model, and Quarto renders the static case study only from validated artifacts.
Goal: Predict future patient visit volumes to optimize staffing and budgeting for a community health center.
- The Challenge: Real patient data is HIPAA-restricted. I needed a way to demonstrate advanced forecasting capabilities without compromising privacy.
- The Solution:
- Synthetic Data Engineering: Wrote Python scripts to generate 5 years of daily clinic data, incorporating realistic seasonality, weekly cycles, and "structural breaks" (e.g., COVID-19 lockdowns).
- Time Series Modeling: Decomposed the data using Statsmodels to isolate trends and applied a SARIMA (Seasonal AutoRegressive Integrated Moving Average) model for 52-week forward predictions.
- Business Impact: Translated model outputs into actionable strategies for dynamic staffing and resilience planning.
- Tech Stack: Python, Pandas, Statsmodels, Scikit-Learn, Seaborn, Quarto.
View the Full Case Study & Forecast
- Languages: Python, R, SQL
- Data Engineering: Pandas, Requests, Pytest, GitHub Actions
- Visualization: Plotly, Seaborn, Tableau
- Modeling: Statsmodels, Scikit-Learn
- Analysis: Time Series Decomposition, Forecasting
- Reporting: Quarto