Hi, I'm George

About

I'm an undergraduate at the University of Michigan pursuing a B.S.E. in Computer Science. I've built production systems as a founding engineer, worked in bank tech, shipped an LLM tool across 26 sites, and published first-author research. Outside of code, I'm into philosophy, game design, bouldering, skateboarding, golf, guitar, and calisthenics.

Experience

Building and scaling production systems.

Capital One

Software Engineer Intern

Jun 2026 – Aug 2026

8 monthsof data staleness eliminated

• Delivered an end-to-end production asset certification feature, replacing manual outreach across 2,000+ assets with a secure PUT endpoint under three-tier authorization and matching React components.
• Built 2 EventBridge-scheduled Lambdas that automate certification reminder emails and daily ServiceNow CMDB sync, eliminating up to 8 months of data staleness from a prior one-time load.
• Authored idempotent Flyway PostgreSQL migrations that add a CMDB caching layer and tune SQL for compliance reporting.
• Configured CI/CD across dev, QA, and prod with Jenkins, IAM roles, SecretsManager, and CloudFormation stacks.

AWS Lambda

EventBridge

Python

PostgreSQL

React

Jenkins

BoilerVault Storage LLC

Founding Engineer · Contract

Jan 2026 – Jun 2026

$130K+revenue reconciled, 0 manual entry

• Reconciled out-of-order payments through webhook pipelines across 4 Stripe accounts and 3 WordPress sites, processing $130K+ in revenue.
• Automated unmatched-charge reconciliation, saving 3+ hours of manual work per week by storing and sweeping oldest-first.
• Built a multi-tenant FastAPI/PostgreSQL/Next.js platform for a 3-campus storage business, with JWT auth across 20+ REST endpoints and a 318-test backend suite.
• Migrated 475 legacy bookings into 2,000+ records across 5 tables, replacing a failing Zapier Google Sheets workflow.

FastAPI

PostgreSQL

Next.js

TypeScript

Stripe

Live Demo

Nexteer Automotive

Software Engineer Intern

May 2025 – Aug 2025

87.5%less code-review time

• Built an LLM-powered IDE extension in Python and TypeScript that parses 300+ internal engineering guidelines and automates compliance checks across C/H files via Azure AI and Claude APIs.
• Tuned prompt pipelines and few-shot strategies to 95% violation-detection accuracy on embedded steering code.
• Deployed to 26 sites, saving engineers an estimated 8 hours of manual code review per week.

Python

TypeScript

Azure AI

LLM

Prompt Engineering

Villanova University

Data Engineer

Jun 2023 – Sep 2023

28,000+pathogen isolates analyzed

• Built R data pipelines that analyze 28,000+ pathogen isolates across 10+ years with PCA and clustering.
• Published the findings as first author in Antibiotics (2023).

R

PCA

Clustering

Data Pipelines

Projects

A mix of systems, ML, and full-stack work. Here are a few I'm proud of.

Architecture

0.482Jul 2026 - Aug 2026

PR-AUC vs. 0.421 baseline

MeloChron

A causal self-attention encoder summarizes a listener's track history and feeds a 468K-parameter 2-layer MLP that predicts whether they return to a newly heard track. Trained in PyTorch on a cloud RTX 4090, cutting a multi-day run to hours while scaling training data 7x. It beat a strong 0.421 PR-AUC baseline to reach 0.482, a gain confirmed by a paired bootstrap significance test over 100,000 listeners.

PyTorch

Transformers

Self-Attention

Recommender Systems

Audio Embeddings

GitHubLive Demo

Preprocessing

0.97Jan 2026 - Apr 2026

validation AUROC

Image Classification & Transfer Learning

A CNN and a from-scratch Vision Transformer, both built in PyTorch, classify dog breeds. A two-stage transfer-learning pipeline pretrains on a 10-class breed task, then fine-tunes a frozen convolutional backbone on a binary task, reaching 0.93 accuracy and 0.97 AUROC.

PyTorch

CNN

Vision Transformer

Transfer Learning

Architecture

3,000+Jan 2026 - Apr 2026

Wikipedia docs indexed

Scalable Search Engine

A multi-stage MapReduce pipeline indexes 3,000+ Wikipedia documents, computing TF-IDF scores and per-document normalization factors across a parallel, multi-job architecture. A Flask API serves 3 partitioned index segments, ranking results by PageRank-weighted cosine similarity, and is deployed to AWS.

Python

MapReduce

Flask

TF-IDF

PageRank

100+Jan 2025 - May 2025

users served

Resume Screener

An NLP pipeline built with TF-IDF and a scikit-learn LinearSVC classifies 12,000+ resumes into 43 categories at 82%/94% top-1/top-3 accuracy. RAG retrieval with FAISS and an LLM ranks resumes against live job postings, deployed via Streamlit to 100+ users.

Python

scikit-learn

LinearSVC

RAG

FAISS

Streamlit

GitHubLive Demo

Skills

Education

Research