I'm Joe Hughes — a software engineer who came up through physics. I build measured, performance-minded software, and I care about the numbers underneath it.
I came to software from physics — and I never stopped thinking like a physicist about it. I want to know why a system behaves the way it does, where it bottlenecks, and how to make it measurably faster.
At Michigan State I trained to reduce complex systems to their governing rules. That same instinct runs through everything I build now: a C++ distributed file engine I benchmarked across sockets, threads, and gRPC; a nutrition platform with a real plan engine doing TDEE math and carb cycling; a glassmorphism control dashboard wired to live telemetry.
Now I'm finishing an MS in Software Engineering at DePaul with a 4.0, interning at Consumers Energy, and shipping side projects faster than I can name them.
Off-screen, same theme: training for the Detroit Marathon, dialing in espresso, keeping a juniper bonsai alive. Measure, adjust, repeat.
A macro & nutrition tracking platform with a personalized plan engine — TDEE calculation, carb cycling, an onboarding wizard, and voice + barcode logging. Built React-first with a dark performance-app aesthetic, designed to scale to iOS and Android.
A four-stage C++ search engine built across a distributed systems course — POSIX sockets, multithreaded indexing with mutex-guarded partitioning, and a gRPC service layer. Benchmarked and tuned for throughput; scored a perfect 50/50 on the autograder across three consecutive assignments.
A custom glassmorphism control surface for a Home Assistant iPad kiosk — hand-built JS cards for hourly weather, live sports scores, and news feeds, plus a Claude-powered voice assistant pipeline and LED strip integration. A real-world exercise in live-data UX.
A cross-platform rewrite of a native Swift/SwiftUI nutrition app onto Flutter — USDA integration, voice logging, barcode scanning, favorites, and a recipe builder. Paired with a multi-agent build pipeline that orchestrates feature work end to end.
A graduate MLOps study of Docker and Kubernetes for machine-learning deployment — covering experiment tracking, model drift, and reproducible pipelines with WandB, MLflow, Hydra, and DVC. Delivered as a formal ACM-format report and recorded presentation.
Before writing code I want the governing rules of the system — the constraints, the data shapes, the real bottleneck. Physics taught me to strip a problem to what actually drives it.
I ship the smallest version that produces a number — a benchmark, a latency, a pass rate. Working software you can measure beats a perfect plan you can't.
Then I iterate against what the numbers say, not what I assumed. Profile, adjust, re-measure. The same loop whether it's gRPC throughput or a marathon split.