Projects & Systems

A PyTorch convolutional neural network for 10-class image classification on the CIFAR-10 dataset.

Designed and trained a residual CNN using convolutional layers, batch normalization, ReLU activations, max pooling, dropout, and shortcut connections. The network contains approximately 3.97 million trainable parameters.

The training pipeline uses random cropping, horizontal flipping, color jitter, random erasing, label smoothing, Adam optimization, weight decay, and scheduled learning rate reductions to improve generalization.

The final model reached 93.12% test accuracy on CIFAR-10. The project also supports classification of individual images and visualization of the 32 activation maps produced by the first convolutional layer.

A campus-only borrowing and lending platform built around the complete lifecycle of lending an item between students.

Built with a four-person Scrum team using Django, Python, PostgreSQL, Bootstrap, and Docker. The application supports item listings, search and filtering, borrowing requests, owner approval, pickup and return tracking, borrowing histories, overdue indicators, and two-way reviews.

Server-side authorization and relational business rules enforce listing ownership, prevent conflicting approved loans, and protect actions based on the user's role in each exchange.

The final release completed all 10 planned user stories and 40 story points with 153 passing automated tests and 98% measured code coverage.

A Java desktop application for securely managing stored credentials through an authenticated graphical interface.

Designed and implemented using an MVC architecture with encrypted persistent storage and controlled credential access.

The JavaFX interface supports authentication and credential creation, retrieval, modification, and deletion while separating interface logic from security-sensitive application behavior.

Input validation and secure data-handling practices were incorporated throughout the application to improve both security and maintainability.

A normalized relational database and Python application for managing theater costumes, props, performers, inventory, and productions.

Designed a PostgreSQL schema using primary, foreign, and composite keys to represent relationships between inventory, performers, shows, and assignments.

SQL queries use joins, aggregation, nested subqueries, and constraint-based filtering for availability tracking and reporting.

A Python command-line interface uses parameterized queries and application-level business logic for inventory operations and availability checks.

My personal portfolio and an ongoing place to document technical work, academic interests, and future computer vision projects.

Designed and developed from scratch using semantic HTML and CSS, with a responsive layout built around typography rather than frameworks or component libraries.

The current redesign reflects my shift toward computer vision and visual computing while keeping software engineering work accessible and easy to explore.