Data Science graduate from Augsburg University (minors in Mathematics and MIS), looking for junior data science, analytics and operations roles. I like work where analysis has to hold up against something real: honest uncertainty, validation against a null or a holdout, and results someone can act on.
Languages: R, Python, SQL, Excel (advanced) Methods: statistical modeling, regression, machine learning, forecasting, Monte Carlo simulation, network analysis, probability, GIS Tools: tidyverse / tidymodels, Shiny, Quarto, FastAPI, ggplot2, Leaflet, ArcGIS Pro, Git
| Project | What it does | Result |
|---|---|---|
| Hennepin County SNAP/MFIP gap analysis | Finds census tracts where eligible residents are not enrolling in benefits, hidden by city-level averages (group capstone) | Random forest: 83.3% accuracy; some tracts above a 90% gap rate inside cities that look fine on average |
| Wake (live app) | Who tends to lead and who follows in a group's music discovery, tested against a permutation null with FDR correction | Interactive app, 4 views; demo shows 94 validated cascades among 3 anonymized listeners |
| Soundings (live app) | Maps a listening history into a taste space (PPMI, 128-d SVD, Leiden communities) | Interactive map of communities that actually co-occur in sessions |
| MLB WAR streaks (report) | Who has the highest WAR in consecutive seasons? Four interpretations across hitters/pitchers, eras, positions and more, with an interactive explorer | 105,912 player-seasons; Walter Johnson's 1912-16 holds the best five-year stretch (69.1 WAR) |
| World Cup Monte Carlo | 5,000 simulated group-stage completions from betting odds | Advance probabilities plus a per-team rooting guide |
| Hall of Fame catchers model | Ridge logistic regression on Lahman data with a time-based split | 0.94 AUC on held-out modern catchers |
More: Listening half-life (how fast a musical obsession fades; median 3 months, interactive charts) · MLB All-Star prediction (11 ML models compared, best F1 0.48) · FF Draft Assistant (Monte Carlo + MCTS live draft tool) · Minneapolis temperature regression · SQL marine rescue database · Sequencium probability analysis · Twin Cities diversity mapping
Procurement analytics in a manufacturing setting. The code and data are proprietary and not shown; these are the headline numbers:
- Modeled about a 73% reduction in purchase-order handling time from consolidating single-line POs, using a full year of PO history (1,385 POs). This is a modeled estimate, not a realized saving.
- Built an automated end-of-life risk report tracking 47 at-risk parts across 4 product BOMs.
- Built an 11-tab spend analytics workbook over 4,968 transactions: 236 suppliers, 1,358 parts and 100 flagged consolidation candidates.
- Built a safety-stock targeting model across about 3,150 BOM components.
- Rebuilt a 21,603-item ABC/XYZ inventory classification from Excel into a documented, config-driven R pipeline, finding and fixing two places where the original model's documentation did not match its formulas.