Projects

Developing an automated computer vision pipeline to track NCAA Division I ice hockey players and transform raw game footage into actionable location data.

I am currently developing an automated computer vision pipeline to track NCAA Division I ice hockey players, transforming raw game footage into actionable location data for advanced team statistics. Last semester, I built the core foundation of this system, achieving a 96.35% tracking accuracy. By evaluating multiple Multi-Object Tracking (MOT) architectures, I successfully addressed the unique challenges of sports environments—specifically severe physical occlusions and top-down camera distortion. My research demonstrated that in visually homogeneous scenarios (like identical team uniforms), spatial state estimation significantly outperforms traditional deep-learning appearance extraction. The current pipeline is fully documented and optimized for scalable cloud-compute (NVIDIA T4 GPU) deployment.

Designed and implemented a novel, dynamically routed hybrid neural network that optimizes resource usage and minimizes noise for Quantum Convolutional Neural Networks (QCNNs).

Recognizing that forcing all features through a quantum circuit is computationally wasteful and prone to errors in the Noisy Intermediate Scale Quantum (NISQ) era, I built a modular pipeline using PyTorch and PennyLane. The core innovation is a trainable router that utilizes a Straight-Through Estimator (STE) to evaluate image features and actively sort them for processing by either a classical network or a Quantum Processing Unit (QPU). To handle scalability issues like barren plateaus, I engineered a classical step-down compression funnel that reduced feature dimensionality before quantum encoding. When deployed and benchmarked on physical hardware via the IBM Quantum System One, the dynamically routed architecture achieved an 82.95% accuracy, demonstrating strong resilience to real-world quantum decoherence and drastically outperforming the pure quantum baseline of 40.34%, proving the architecture's viability as an effective noise mitigation strategy.

Engineered an end-to-end quantitative machine learning pipeline to predict demand shocks in CME Lumber Futures following major hurricanes.

We engineered an end-to-end quantitative machine learning pipeline to predict demand shocks in CME Lumber Futures (LBS=F) following major Atlantic hurricanes. Our architecture maps raw physical meteorology into actionable trading signals using a two-step model to isolate the 'Wave 2 Reconstruction' period: The Physics Engine (a deep LSTM neural network forecasting coastal storm surge across 7 critical East Coast timber ports using live NOAA/NDBC sensor data) and The Financial Regressor (a Random Forest classifier weighing physical damage against live macroeconomic liquidity, specifically S&P 500 momentum and 10-Year Treasury Yields). In backtesting, the pipeline achieved an 83.3% precision rate in identifying historical >5% bullish price spikes over a 9-month forward window.

I led the delivery of a production-ready Windows utility integrated into Microsoft PowerToys.

• Led a cross-functional team delivering a production-ready Keystroke Overlay utility for Microsoft PowerToys.
• Coordinated Agile sprints, client meetings, and stakeholder communications with Microsoft product managers.
• Managed scope, timelines, and final knowledge transfer to ensure a successful client handoff.
• Conducted financial modeling and ROI analysis to demonstrate strategic value and adoption impact.

Keystroke Overlay was a semester-long capstone project completed in collaboration with the Microsoft PowerToys team. The project delivered a fully integrated Windows utility that visualizes keyboard input in real time, addressing a highly requested community feature for presenters, educators, and accessibility users. I served as Project Manager, overseeing all non-coding aspects of the project, including Agile planning, stakeholder communication, and scope management. I also conducted financial modeling and ROI analysis to articulate the project’s strategic value to Microsoft. The project concluded with a successful client review, handoff, and integration into the PowerToys ecosystem.

I built a web application teaching budgeting skills through interactive exploration.

• Earned 3rd place at the January 2024 NSBE–SHPE Hackathon with a team of four.
• Designed and developed a web app teaching budgeting skills through interactive city exploration.
• Built the frontend using React, integrating Mapbox for geospatial interaction.
• Presented technical design and functionality to judges and attendees.

Trip to Troy was a web-based project developed during the January 2024 NSBE–SHPE Hackathon, where our team earned 3rd place overall. The application teaches essential budgeting skills while highlighting small, minority-owned businesses in the local community, aligning with the event’s focus on financial literacy and underserved groups. I developed the frontend using React and integrated Mapbox to support interactive geographic exploration. I also presented the technical implementation and design decisions to judges, strengthening my experience in rapid prototyping and technical communication.

I am building a modern, performant personal website to showcase my experience and projects.

• Designed and developed a personal portfolio website using modern web technologies.
• Implemented reusable, data-driven project and experience components for scalability.
• Focused on performance, accessibility, and responsive design across devices.
• Iterated on UI/UX and architecture to support future expansion and content growth.

This project is an ongoing effort to design and build a modern personal website that showcases my professional experience, projects, and technical interests. I am developing the site with an emphasis on clean architecture, reusable components, and maintainability, allowing experiences and projects to be easily extended over time. The site is designed to be fully responsive and accessible, with careful attention to performance and user experience. Beyond serving as a portfolio, the project acts as a sandbox for experimenting with new tools, frameworks, and design patterns while maintaining production-quality standards.