Undergraduate at the University of Southern Mississippi interested in
machine learning theory, mechanistic interpretability, distribution shift, and mathematical foundations of ML.
I am a Computer Science and Mathematics student interested in understanding machine learning systems from both mathematical and computational perspectives.
My current interests include:
- Mechanistic Interpretability
- Machine Learning Theory
- Distribution Shift & Robustness
- Machine Unlearning
- Mathematical Reasoning in Language Models
- Optimal Transport
- Probabilistic Machine Learning
I particularly enjoy projects where mathematical ideas can be turned into reproducible experiments.
A reproducible benchmark investigating whether label-free distribution-distance measures can provide early warning of classifier degradation.
The project compares:
- Wasserstein distance
- Sliced Wasserstein distance
- Maximum Mean Discrepancy
- KL divergence
- Population Stability Index
across multiple datasets, classifiers, seeds, and shift mechanisms.
Stack: Python · NumPy · pandas · SciPy · scikit-learn · Matplotlib · pytest
A mathematical implementation of Bayesian linear regression covering:
- Gaussian likelihoods and priors
- MAP estimation
- Posterior inference
- Predictive uncertainty
- Regularization from a Bayesian perspective
Includes mathematical derivations alongside the implementation.
Stack: Python · NumPy · scikit-learn · Matplotlib · Jupyter
Study of Principal Component Analysis from the maximum-variance and SVD perspectives.
Includes:
- dimensionality reduction
- eigen-digit visualization
- image reconstruction
- residual analysis
- mathematical derivations
Full-stack household expense-sharing application with authentication, shared expenses, tax calculations, and REST APIs.
Stack: React · TypeScript · Vite · Django · Django REST Framework · JWT
Machine-learning analysis of historical Formula 1 data with reusable feature engineering and modeling pipelines.
Includes:
- race and pit-stop analysis
- pre-race feature engineering
- finish-position regression
- podium classification
- saved preprocessing/model pipelines
Stack: Python · pandas · scikit-learn · Jupyter
Budgeting application with persistent storage, authentication, transaction management, budgeting, statistics, and containerized execution.
Stack: Python · Kivy · SQLite · Docker
Mechanistic Interpretability
├── Transformer Circuits
├── Activation & Path Patching
├── Patchscopes
├── Residual Stream Representations
└── Machine Unlearning
Mathematical ML
├── Probability
├── Real Analysis
├── Bayesian Inference
├── Optimal Transport
└── Distribution Shift
Languages
Python · C++ · Java · JavaScript · TypeScript · SQL
Machine Learning & Data
NumPy · pandas · scikit-learn · Matplotlib · PyTorch · Jupyter
Backend
Django · Django REST Framework · FastAPI · REST APIs
Frontend
React · Vite · TypeScript
Tools
Git · GitHub · Linux · Docker
I am particularly interested in questions such as:
How are computations represented internally in neural networks?
How can we detect when machine-learning systems encounter distributions different from those on which they were trained?
Can model representations be understood or modified precisely enough to enable reliable interpretability and machine unlearning?
My long-term goal is to work on mathematically grounded problems in machine learning and AI research.
