Patrick Cowhill

Build Difficult Software

Hands-on implementation across Java, C++, Python, web systems, cloud infrastructure, testing, integration, and deployment.

Understand the Problem Deeply

A foundation in mathematics, physics, electrical/computer engineering, modeling, signal processing, and research.

Take Software to Real Users

Requirements, customer interaction, demos, deployment, feedback, debugging, and iteration.

Make Teams More Effective

Technical leadership, mentoring, teaching, code review, presentations, and coordinating work without stepping away from engineering.

Roster5

April 2026 – Present

Software Engineer

Java
Spring Boot
TypeScript
Cypress
AWS
Kubernetes
DynamoDB
MongoDB

Hands-on engineering on production programs

I joined Roster5 in April 2026 as a hands-on software engineer. Day to day, I am writing and shipping code on the company's programs: Java and Spring Boot services, TypeScript and modern web front ends, Cypress and other automated tests, and deployments on AWS and Kubernetes backed by DynamoDB and MongoDB.

Most of that work so far has been what a new engineer on a mature program should be doing: picking up tickets, learning the systems and their history, fixing real problems, and getting changes tested and into production while coming up to speed on the team's technologies.

AI as part of everyday engineering

AI-assisted development is part of my normal workflow, and I have become one of the internal technical voices helping Roster5 work out how to use it effectively and responsibly:

Delivered roughly four internal presentations to the Roster5 team on practical AI-assisted software development, including project demonstrations, usage guidance, and concrete examples of AI in engineering workflows.
Consulted directly with the company president and vice president on AI adoption, responsible use, and opportunities to incorporate AI into upcoming contract work.
Presented to customers on one of Roster5's programs about how the company is using AI in software development.
Own and administer Roster5's Claude Team subscription.

L3Harris Technologies

January 2020 – April 2026

Associate Software Engineer
Senior Associate Software Engineer
Specialist Software Engineer

C++
Python
MATLAB
Linux
GPU acceleration
Containers
GovCloud

From image-processing algorithms to deployed geospatial systems

Over six years at L3Harris I progressed from Associate to Specialist Software Engineer while staying on the implementation side of the work. I built electro-optical (EO) and synthetic aperture radar (SAR) software in C++, Python, and MATLAB on Linux, including GPU-accelerated image processing, signal processing, and geospatial systems.

The job ran the full lifecycle: turning customer requirements into designs, writing and reviewing code, debugging hard problems, packaging software in containers, integrating and testing it, managing configuration, and deploying to both GovCloud and on-premises environments. I regularly demonstrated working software to stakeholders and customers.

As the scope grew, so did the responsibility around the code. I led engineering efforts involving up to 20 people and served as product owner for three deployed solutions, while continuing to write software on those same systems. That combination of technical demos, customer engagements, and proposal work supported roughly $4M in annual program revenue.

Issued U.S. patents (listed as inventor)

U.S. Patent No. 11,636,649 B2 Geospatial System for 3D Model Updates Based Upon Predictively Registered Image
U.S. Patent No. 11,816,793 B2 System for 3D Geospatial Model Updates via Iterative Predictive Image Registration

Schematic of a scan pipeline: a queue of more than one million websites flows through parallel TLS probes into a results store and analysis stage.

Case Study · Georgia Tech

Network Security Study: HTTPS/TLS Adoption at Scale

Problem. How widely, and how well, is HTTPS/TLS actually deployed across the web, and what does the long tail of sites look like?

For a graduate network security project at Georgia Tech, I queried and analyzed more than one million websites to examine how HTTPS and SSL/TLS certificates are adopted and implemented by online services hosted in China. That meant building scripted, automated data collection that could run reliably at scale, storing and cleaning the results, analyzing adoption and configuration patterns, and then interpreting what the data did and did not say about the implications for both servers and clients. The findings were written up as a paper and presented.

What this demonstrates: technical research, automation at scale, network security, data analysis, and technical communication.

Abstract route map of cities connected by colored track segments, with several agents' claimed routes highlighted and a small results histogram.

Case Study · Modeling & Simulation

Ticket to Ride Simulation

Problem. Which strategies actually win in a rules-heavy board game, and how do you find out without playing thousands of games by hand?

Working with a team in MATLAB, I helped translate the full rule set of Ticket to Ride into a faithful simulation: train pieces, cards, routes, turns, and scoring. We then programmed artificial agents with distinct decision-making processes, ran hundreds of simulated games in which they competed against one another, and analyzed the performance metrics to identify the most successful, reliable, and complementary strategies. The project was as much about clear collaborative software design as it was about the algorithms.

What this demonstrates: modeling, simulation, algorithm design, collaborative software development, and translating complex systems into software.

Flow diagram of a GitHub-centered delivery loop: requirements to issues, implementation, pull request, independent review, automated tests and CI/CD, deployment, and feedback back into issues.

Case Study · Modern Engineering

GitHub as the Operating Environment for an AI-Assisted Software Team

Problem. Can the ordinary machinery of professional software delivery keep an AI-driven team honest, reviewable, and shippable?

In a recent experiment I used GitHub itself as the operating environment for an AI-driven software team. A succession of independent Claude Code sessions act as product manager, developer, reviewer, QA engineer, release engineer, and maintainer, each orienting itself entirely from the repository, its issues, pull requests, review comments, and CI results. Requirements arrive as issues from a human customer; implementation lands through pull requests; every change goes through independent review, automated unit and browser tests, and lint and type checks; CI/CD deploys the result; and feedback on the deployed product feeds back into new issues. The product that emerged is a browser-based video editor. The point was not to remove engineering discipline but to show that the same Git-based workflow, tests, and review gates that make human teams reliable make AI-assisted engineering reliable too.

What this demonstrates: modern software delivery, AI-assisted engineering, Git-based workflows, automated testing, CI/CD, and iterative development.

Georgia Institute of Technology

M.S. in Electrical & Computer Engineering

GPA 4.0

Completed while working full-time at L3Harris and taking on substantial professional engineering responsibility.

Areas of study

Computer architecture
Signal processing
Adaptive filtering
Network security
Mathematics of machine learning
Modeling and simulation
Computer vision
Partial differential equations
Advanced programming

University of Florida

B.S., majors in Physics and Mathematics

GPA 3.95

The source of my analytical background, mathematical reasoning, modeling mindset, and scientific research experience.

Undergraduate biophysics research with MATLAB-based modeling.
Guest lectured in upper-level Mechanics II, using Mathematica to teach computational approaches to physics problems.

Own the problem, not just the ticket: Understand the need, constraints, users, and tradeoffs before converging on a solution.
Build, test, and iterate: Prefer working software, feedback loops, automated testing, and incremental improvement.
Communicate clearly: Explain technical decisions appropriately to engineers, customers, leadership, and other stakeholders.
Help the team improve: Share knowledge, mentor others, review work thoughtfully, teach new tools, and improve engineering practices without losing touch with implementation.

Explore my experiments, tools, games, prototypes, and personal projects at cowhill.dev.

Visit cowhill.dev

Engineering across the whole problem, not just one layer of it.