Academics //
Franklin & Marshall College
Academics
Research interests, coursework, and research projects from my Computer Science degree and my work since.
B.A. Computer Science · GPA 3.72 · Magna Cum Laude
Research interests
Honest model evaluation when data is noisy, biased, or shifting. In my projects I compare random cross-validation with forward-in-time tests and held-out regions (Melbourne housing: 13.1% error on random folds versus 18–38% on an unseen region), and build the data pipelines that make those tests reproducible.
Education
Franklin & Marshall College, Lancaster, PA
Bachelor of Arts in Computer Science, May 2026
GPA 3.72
Magna Cum Laude. Honors List (5 semesters). Dean's List (Spring 2022).
Coursework
Computer Science (transcript grades)
- Intro to Machine Learning · B+
- Artificial Intelligence · A-
- Algorithms · A
- Data Structures (C++)
- Theory of Computer Science · B+
- Information Theory · A-
- Computational Discrete Math · A
- Computer Organization · A
Mathematics & Statistics (transcript grades)
- Applied Regression Analysis · A
- Probability and Statistics I · A-
- Intro to Statistical Modeling · A
- Intro to Data Science · A
- Calculus II · A
- Calculus III · B
- Linear Algebra and Differential Equations · Pass (no letter grade)
Independent projects
Can Mail Data Predict Housing Vacancy?
Microsoft Fabric, Power BI. Sep 2026
- Investigated whether quarterly USPS vacancy counts can stand in for the annual Census ACS vacancy rate, building a reproducible pipeline over 6 counties (84 usable county-year observations).
- Evaluated the hypothesis three ways (correlation, directional accuracy, bias): a pooled r = 0.77 fell to between −0.01 and 0.72 within counties, with only 57.7% directional agreement, so USPS data is a county-dependent signal at best, not a substitute.
- Flagged a 2012 USPS classification change as a data artifact and fixed two silent ingestion bugs, preventing a false trend.
Grocery Transaction Data Warehouse
SQL Server, T-SQL. Jul–Aug 2025
- Built a Medallion-architecture warehouse over 131M+ transaction lines with stored-procedure ETL, a star-schema Gold layer, and 21 SQL data-quality checks; found that 47.5% of completed orders had totals that did not match their line items, and rebuilt revenue from line prices.
Medical Cost Modeling and Prediction
Python, scikit-learn, R. Oct 2026
- Extended my 2021 R study of 1,338 insurance records: a smoking × obesity interaction is associated with the three cost bands, raising in-sample R² from 0.75 to 0.86 and showing the earlier sex effect was confounded by smoking (it flips from +$1,539 to −$494).
- Compared 9 models with 5-fold CV; the interpretable linear model matched ridge, lasso, and random forest and beat boosting (held-out R² 0.88), but subgroup analysis found 29% relative error for non-smokers versus 5% for obese smokers.
Melbourne Housing Price Prediction
Python, scikit-learn. 2026
- Tested four models on 13,580 Melbourne home sales three ways: gradient boosting’s error was 13.1% under random 5-fold CV and also held on future sales, but rose to 18–38% on regions held out entirely.
Can a Model Nowcast Census Vacancy from Mail Data?
Python, NumPy, scikit-learn, SQLite. Oct 2026
- Compared eight models under three evaluation designs; none beat carrying last year’s Census value forward in a realistic forward-chaining nowcast (MAE 1.023 vs 1.054 for ridge; with only 78 rows, I cannot say whether that gap is real), although two standard schemes would have shown a small win.
- Wrote gradient-descent linear and logistic regression from scratch with tests against closed-form and scikit-learn results, and found 80% conformal intervals that cover 81% pooled but only 57% in one county.
AI, Systems & Theory Coursework
Python, C++, RISC-V. 2024 – 2026
- Implemented uninformed and informed search, constraint satisfaction, logical inference, and Bayesian networks in Python, and solved a Connect-4 board puzzle with state-space search.
- Implemented k-means and gradient-descent linear and logistic regression from scratch in Machine Learning (Steam games; 13.6K Melbourne housing listings), and later re-tested the housing data with gradient boosting on held-out regions (error 13% → 18–38%).
- Designed a CPU in Logisim, wrote recursive RISC-V assembly, coded DFAs and NFAs, and benchmarked a red-black tree in C++.
Certifications
- Microsoft Certified: Fabric Data Engineer AssociateMicrosoft, Oct 2026. Credential ID 5CE20E8CF4AC4CE3
- SQL (Advanced) CertificateHackerRank, Sep 2025
Leadership
Technology Committee, Diplomatic Congress
Franklin & Marshall College. Aug 2021 – Aug 2022
Cut classroom technology costs by 18% by leading IT policy work and centralized license management across 15+ departments.
Service
Service Year Participant, The Shalom Project
Lancaster, PA. Aug 2026 – Jul 2027
Full-time service year placed at the Center for Regional Analysis, living in an intentional community with a cohort of young adults.