Selected work

Projects

A few things I've built, from starting Michigan's first eVTOL team to co-authoring ML research on space weather. Each one is real, runs on actual data or hardware, and is still moving.

U-M Vertical Flight

Role: Founder & Electrical Lead
Period: Jan 2026 – Present
Status: Active

Vertical Flight Society

Michigan's first eVTOL design team, competing in Vertical Flight Society design-build-fly events.

U-M Vertical Flight is a student design team I started to compete in Vertical Flight Society design-build-fly competitions — national events judged on technical design reports and live flight performance. As Founder and Electrical Lead, I own the full electrical stack on a real flying aircraft, from power architecture to sensor integration.

What I did

Founded the team from scratch; recruited a multidisciplinary group of engineers and secured $20k+ in funding for hardware, components, and competition entry.
Design and integrate the full electrical system: power distribution, flight controllers, ESC wiring, and sensor suites (LiDAR, thermal cameras) for an eVTOL UAV.
Coordinate CAD in SolidWorks and Siemens NX across mechanical, electrical, and software sub-teams to meet VFS competition design specifications.
Lead team operations: sprint planning, documentation, budget management, and presentations to faculty advisors and industry sponsors.

Stack

Electrical SystemsSolidWorksSiemens NXC++LiDARLeadership

Geomagnetic Event Prediction

Role: Research Assistant, U-M Magnetometer Laboratory
Period: Aug 2025 – Present
Status: Active

View on GitHub

The first ML classifier to distinguish Storm Sudden Commencements from Sudden Impulses without upstream solar wind data.

Classifying a geomagnetic event today takes a 48-hour observation window after the initial spike to confirm whether a storm actually follows — far too slow for real space weather forecasting. Our Random Forest model classifies the event type in real time using only ground-based Sym-H magnetometer features, so it no longer depends on L1 solar wind satellites like DSCOVR and ACE. I'm co-authoring the paper.

What I did

Trained Random Forest classifiers on 20 years of Ebre Observatory SC events (2006–2025, a 3.7:1 SSC-to-SI ratio spanning Solar Cycle 24), reaching 80% recall on the minority class.
Engineered physically grounded features from 1-minute Sym-H data: 24-hour pre-event standard deviation (geomagnetic quietness), 10-minute maximum amplitude jump (compression shock), and an amplitude-to-variability ratio.
Validated with Leave-One-Year-Out cross-validation across Solar Cycle 24 phases to characterize minimum data requirements for operational deployment.
Conducted multi-station superposed epoch analysis with the SuperMAG network to confirm the physical basis of the engineered features by latitude.
Calibrate the lab's Helmholtz Coil apparatus for RM3100 magnetometer testing, validating hardware reliability for spaceflight environments.

Stack

PythonScikit-learnRandom ForestMATLABSuperMAGSym-H

GoFundMI.org

Role: Non-Profit Co-founder
Period: Feb 2026 – Present
Status: Active

In development — public launch to come

Connecting student researchers with the funding their work deserves.

Most research funding friction is really an information problem: donors don't know which projects exist, and students don't know which donors are looking. GoFundMI is a non-profit platform that gives donors a direct line to student-led university research. I'm building it with Curtis Ling, founder and CTO of MaxLinear. The platform is in active development.

What I did

Co-designing the platform architecture in Python and React to serve the University of Michigan research community.
Building transparent, direct funding pipelines between donors and student-led research projects — profiles, milestones, and verified outcomes.

Stack

PythonReactWeb DevelopmentNon-Profit

Summa-Fi

Role: Personal Project
Period: Sep 2025 – Present
Status: Active

View on GitHub

Secure, self-hosted financial intelligence.

A self-hosted financial platform that classifies transactions automatically, pulls real-time data through the Plaid API, and visualizes everything in React. It cut my manual accounting work by about 70% — and because it's self-hosted, my financial data never leaves my own infrastructure.

Demo

What I did

Architected Python backend services that ingest and classify financial events with explainable outcomes, processing thousands of transactions.
Integrated the Plaid API for real-time data ingestion, PostgreSQL for persistent storage, and React for interactive visualization.
Built workflow automations that reduced manual accounting time by ~70% and provide real-time visibility into spending patterns.

Stack

PythonReactPostgreSQLPlaid API

Curious about what's next? A few ideas are waiting on the drawing board.

Ideas Contact