Experience
Data Analyst Intern
Summer 2026- Authored a data-rules specification defining about 25 client and utilization metrics, resolving a cross-team cancellation-rate discrepancy so management could track performance using consistent measures.
- Proposed and completed a competing-risks analysis of 2,121 clients using cause-specific Cox models, with time-varying and influence diagnostics. A 10-percentage-point higher Stage-1 attendance rate was associated with lower dropout hazard at day 90 (HR 0.840) and day 365 (HR 0.936), providing a basis to test early attendance support as a retention strategy.
- Built a 39-chart Mode report from 23,406 client-week observations in PostgreSQL; decomposed a 313-hour monthly decline and found 78.4% remained outside client-count and recorded-cancellation drivers, showing those factors were insufficient explanations.
Projects
U.S. Unemployment Rate Forecasting
R · R Markdown
Built a reproducible forecasting study on 942 observed monthly BLS unemployment rates (1948–2026), preserving a source-documented missing month and reserving the final 24 calendar months as holdout. Compared 24 diagnostically specified SARIMA models and ETS across 10 rolling windows; on a held-out 24-month period, the selected model beat naive and drift on MAE but not RMSE, exposing regime sensitivity. The analysis also reports interval and residual diagnostics.
Chaotic Attractors: Computational Exploration & Visualization
Python · NumPy · SciPy · Matplotlib
Built a pip-installable Python library (CLI + API) for searching, validating, and visualizing 9 four-parameter 2D systems (5 established forms, 4 custom variants). Designed a seeded, staged search that removes burn-in and validates survivors with finite-time largest Lyapunov estimates; ranked outputs by geometry and rendered density with subsampled Gaussian KDE. Includes PNG/PDF/SVG export.
USB Watchdog: macOS Hardware-Tamper Monitor
Python · Bash · C · IOKit
Built a macOS menu-bar tamper alarm with a Python GUI and Bash engine that monitors USB, Thunderbolt, SD-card, and external-display inventories and triggers either a normal shutdown with forced-halt fallback or an immediate forced halt upon an observed change. Implemented a native IOKit event listener with an independent polling fallback; regression-tested event failover, bounded probe failures, wake-time revalidation, process identity, and both shutdown policies, then packaged the system as a reproducible app bundle. The system is a physical-change tripwire, not device authentication.