Portfolio - Longevity & Healthcare Analytics
Longevity and healthcare data analyst. I turn aging, biomarker, and clinical data into decision-ready insight, pairing real bioinformatics work with 15+ years of operations and analytics experience. Recent work includes a DNA methylation aging clock that predicts biological age to within 5.65 years on held-out blood (Pearson r 0.892) and holds at 6.36 years on an independent cohort. A companion mortality study on 28,510 NHANES adults found each year of biological-age acceleration carries 4.5 percent higher mortality risk, independent of chronological age.
Filter by skill
Aging clocks, biomarker modeling, and gene-expression analysis on public omics and clinical data.
A live dashboard for the methylation aging clock. Move the sliders on the gene-labeled CpG sites the model weights most and the predicted biological age updates in real time. Held-out accuracy of 5.65 years (Pearson r 0.892), holding at 6.36 years on an independent 689-person cohort, the honest cross-study number.
An interactive analysis of biological age and survival in 28,510 US adults from NHANES (1999-2010), with mortality follow-up through 2019 (6,265 deaths, median 12.9 years). Implements the Levine PhenoAge score from nine routine blood biomarkers and tests it against all-cause mortality with Cox regression and Kaplan-Meier survival. Each year of PhenoAge acceleration was associated with 4.5% higher mortality, independent of chronological age (HR 1.044, 95% CI 1.043 to 1.047), and PhenoAge out-discriminated chronological age (C-index 0.865 vs 0.848). Filter by age and sex and explore survival curves and biomarker distributions.
An interactive explorer for genes differentially expressed between young and old human cortex (GSE104704). A DESeq2 pipeline found 119 genes significant at 5% FDR, led by myelin and oligodendrocyte genes. Search any gene for a plain-language readout of its age-related change plus a live NCBI summary, explore an interactive volcano, and filter the full results. Includes MA plot and pathway enrichment.
A searchable tool that grades popular longevity interventions by how strong the evidence actually is, using a transparent six-axis rubric, and visualizes the gap between public attention and real evidence. Grade A requires a human lifespan trial, which nothing currently earns. Citations resolve live from PubMed, and every grade is explained in plain language.
Predicts biological age from whole-blood DNA methylation. Streams GSE40279 (656 samples, about 450,000 CpG sites) with a memory-safe loader and trains a cross-validated elastic-net to 5.65 years (r 0.892) using a sparse 176-CpG signature it selected itself. An enrichment test confirms the selected sites overlap established Hannum blood-clock CpGs (hypergeometric p = 2.7e-06).
An interactive analysis of how mouse lung ages, cell by cell (Tabula Muris Senis). Across 11 lung cell types it finds composition shifts and age-associated genes using pseudobulk differential expression per mouse, the statistically correct approach that avoids the pseudoreplication that invalidates per-cell tests. Includes UMAP, composition, and senescence scoring.
Clinical risk modeling, plus the same 300-patient readmission analysis carried across R, Power BI, Python, Looker, Sheets, and Excel.
A 30-day hospital readmission risk analysis on a 300-patient clinical dataset, executed across R, SQL, Power BI, and Excel to show the result holds across tools. The R logistic regression (glm) quantifies how each factor shifts the odds: relative to surgical patients, Heart Failure carries roughly 20x the odds (OR 19.7, p<0.001), with COPD, Sepsis, and Pneumonia all 6-10x and significant, reported as odds ratios with 95% confidence intervals and ROC AUC 0.714.
Interactive dashboard analyzing readmission patterns across 300 patients and 5 diagnoses. Heart Failure patients showed a 39% readmission rate, nearly 6x higher than surgical cases. Built with DAX measures and interactive slicers filtering by insurance type, department, and gender.
Machine learning pipeline predicting 30-day hospital readmission risk from patient records. Applies feature engineering, model training, and evaluation to identify high-risk patients before discharge. Built end-to-end in Python with scikit-learn.
Interactive Looker Studio dashboard analyzing 300-patient hospital readmission dataset. Features scorecards, bar charts by diagnosis and insurance type, pie chart by discharge destination, and a live dropdown filter. Connects directly to Google Sheets as the data source.
Advanced spreadsheet analysis using 3 pivot tables, COUNTIFS, COUNTIF, INDEX/MATCH, and calculated readmission rates by diagnosis and insurance type. Dashboard tab pulls live KPIs from formula sheet. Heart Failure readmission rate 39.29% vs 7.04% for Hip/Knee Replacement.
Native Excel workbook analyzing the 300-patient readmission dataset. Three pivot-style summary tables (by diagnosis, insurance type, and discharge disposition) built with COUNTIFS and AVERAGEIF, plus an INDEX/MATCH lookup returning the highest-risk diagnosis. Dashboard tab links live to the formula sheet via cross-sheet references with a readmission-rate bar chart. Patients without a scheduled followup readmit at 30.59% vs 24.00% overall.
Python engineered a weighted attrition risk score per employee across 8 factors. A Power BI dashboard on the enriched dataset identifies Sales and Marketing at 28% attrition, 8 points above the 15% target, with a scatter plot, department treemap, gauge, and interactive slicers.
Full-stack web applications shipped to production.
Full-stack Flask web app shipped to production that generates shareable Destiny 2 loadout links by integrating the Bungie.net and Destiny Item Manager APIs end to end, with a server-side OAuth 2.0 token exchange. Diagnosed and fixed live production failures including Cloudflare bot protection and platform-membership resolution. Deployed on Render.
Full-stack Flask web application shipped to production with a dual-edition architecture: an online optimizer on Render and a desktop edition with live market pricing via REST API. Managed a full open-source release pipeline with GitHub CLI versioned releases and a downloadable desktop package.
SQL, BI dashboards, and earlier exploratory and practice work.
6 queries across 4 real PostgreSQL databases demonstrating CTEs, window functions, conditional aggregation, subqueries, and date functions. Key finding: 210 whales generate 52% of game revenue despite being 9% of paying users.
Interactive Tableau dashboard analyzing 2019 World Happiness Report data across 156 countries, with a choropleth world map, GDP versus happiness scatter with trend line, top 20 bar chart, and factor comparison.
End-to-end web analytics project exploring Grammy Awards data. Applies data wrangling, exploratory analysis, and visualization to uncover audience and engagement patterns.
Hands-on Python lab exercises covering data manipulation, functions, and analysis workflows. Demonstrates foundational programming competency applied to real datasets.
9-part project series covering progressively advanced analytics techniques from data cleaning and aggregation through statistical analysis and visualization. Datasets include Dunder Mifflin sales, Nike/Adidas, and Lyft bike share data.