Available for opportunities · Denton, TX

Alex
Metzger

Full-Stack Developer & AI Researcher

I build full-stack web applications and AI-powered tools — from production e-commerce sites to conversational AI integrations. Currently a PhD Student at the University of North Texas with a completed M.S. in Artificial Intelligence and two papers published at CogMI 2025.

View Projects ★ Publications Download Résumé ↗
GitHub LinkedIn Email 575-805-9738
Portrait of Alex Metzger
PhD Student AI × Cybersecurity Lab

Who I am.

4.0
Master's GPA
4
Shipped Projects
2
Research Papers

My technical background spans backend development with Flask and Node.js, full-stack web applications, and AI/ML engineering. As a PhD Student at UNT, I research transformer-based LLMs, ML pipelines, and prompt engineering — with two papers on multi-label classification and cybersecurity AI published at CogMI 2025.

Alongside my academic work, I run The Domain Designers — a web design studio where I work directly with real clients to design and build custom websites under real-world constraints. I'm also genuinely excited about vibe coding as a discipline — actively learning AI-assisted development workflows while staying grounded in the fundamentals that make the output worth anything: clean architecture, readable code, and sound engineering practices.

I hold a B.S. in Computer Science from Southern New Hampshire University (Magna Cum Laude, 3.73 GPA) and a completed M.S. in Artificial Intelligence from UNT (4.0 GPA), where I'm continuing as a PhD Student.

Full Resume ↗

// research output

Peer-Reviewed
Publications.

3
papers
published
CogMI 2025 Pittsburgh, PA
Dependence Minimization for Multi-Label Classification: An Alternative to Human Labeling
↗ DOI: 10.1109/CogMI67134.2025.00027
CogMI 2025 Pittsburgh, PA
Prompts and Thoughts: Can Your Cyber Curriculum Meet the Job Skills
↗ DOI: 10.1109/CogMI67134.2025.00039
IntelliSys 2026 Amsterdam, Netherlands
Task-Aligned Contrastive Learning for Filtering Noise in Multi-Label Text Classification
// pending DOI
// 2 IEEE · 1 pending

Selected Work

A handful of things I've built. Each one is a real system, not a demo.

01
PythonFlaskRedisOpenAI APIJavaScript

JukeboxRadio AI

An AI-powered internet radio application where users discover and request music through a conversational interface. Built on Flask with Redis managing real-time session state and listener queues, it integrates the OpenAI API alongside MusicGPT to interpret natural language requests and serve curated streams. The core challenge was orchestrating multiple AI APIs under a low-latency constraint while keeping concurrent listener sessions isolated and responsive.

deployed · jukeboxradioai.com
Live Site → GitHub ↗
02
PythonFlaskMySQLAIJavaScript

OmniClip

A full-stack AI application for capturing, organizing, and retrieving content. Flask handles the backend with MySQL for persistent storage, and an AI layer processes incoming content to make it automatically tagged and searchable. The design goal was reducing the friction of content management — building a system where retrieval is intelligent enough that users don't have to think carefully about how they organize things when they save them.

GitHub ↗
03
PHPMySQLStripe APIJavaScriptHTML5/CSS3

Southwest Candles

A fully custom e-commerce website built for a real client — no platform, no shortcuts. The stack handles everything a production store needs: a MySQL-backed product catalog, Stripe payment processing, order management, and a responsive storefront. Every piece of the purchase flow, from cart state to post-payment confirmation, was implemented from scratch and has been in production use since launch.

deployed · southwestcandles.shop
Live Site → GitHub ↗
04
JavaScriptPHPReact.jsNode.jsHTML5/CSS3

The Domain Designers

The web design studio I founded and run. Work here is client-focused: real briefs, real deadlines, and real expectations from business owners who need a site that actually serves their goals. Managing client relationships means navigating shifting requirements, communicating tradeoffs clearly, and delivering on time when there's no room for over-engineering. Running this alongside a PhD has made me a sharper estimator, a cleaner communicator, and a more pragmatic developer.

deployed · thedomaindesigners.com
Visit Studio →

Skills & Toolkit

Technologies and domains I work in regularly.

GitHub Activity · Metameg

Profile · @Metameg
—public repos
—last push
—followers
—active this yr
on github since —
Top Languages
loading…

Languages

  • Python
  • JavaScript
  • PHP
  • SQL
  • Bash

AI & Machine Learning

  • PyTorch & TensorFlow
  • Hugging Face Transformers
  • OpenAI API
  • LangChain
  • Prompt Engineering
  • Scikit-learn

Web & Backend

  • Flask & Node.js
  • React.js
  • MySQL & Redis
  • Stripe API
  • RESTful APIs
  • Docker

Research & Expertise

  • Multi-label Classification
  • Cybersecurity AI
  • Natural Language Processing
  • LLM Fine-tuning
  • ML Pipeline Design
  • Published Research (CogMI 2025)
Currently
PhD Student · University of North Texas Seeking full-time developer roles

Let's build something
worth building.

Open to full-time roles, internships, and research collaborations. I read every message — don't hesitate to reach out.

./send_message.sh

Email ametzger08@gmail.com
Phone 575-805-9738
LinkedIn linkedin.com/in/alex-metzger-699802233
GitHub github.com/Metameg
Résumé Download PDF
Open to
Full-time Developer roles Freelance projects Research collaborations

★ publications/

Research & Publications.

Peer-reviewed work in multi-label classification, NLP, and cybersecurity AI.

01 2025
CogMI 2025 Pittsburgh, PA

Dependence Minimization for Multi-Label Classification: An Alternative to Human Labeling

We propose a dependence minimization framework that eliminates the need for manual human-annotated label co-occurrence data in multi-label classification. By leveraging statistical independence criteria applied directly to observed label distributions, the method learns inter-label relationships from raw co-occurrence structure — reducing annotation cost while preserving classification performance across benchmark datasets.

DOI: 10.1109/CogMI67134.2025.00027 ↗ Read Paper →
02 2025
CogMI 2025 Pittsburgh, PA

Prompts and Thoughts: Can Your Cyber Curriculum Meet the Job Skills

We benchmark current cybersecurity degree curricula against real-world LLM-era job postings using prompt engineering techniques applied to large language models. Our analysis identifies critical skill gaps between what programs teach and what employers now demand — particularly in AI-assisted threat analysis, prompt-injection awareness, and model security — offering actionable recommendations for curriculum modernization.

DOI: 10.1109/CogMI67134.2025.00039 ↗ Read Paper →
03 2026
IntelliSys 2026 Amsterdam, Netherlands

Task-Aligned Contrastive Learning for Filtering Noise in Multi-Label Text Classification

Proposes an unsupervised contrastive fine-tuning approach applied to an LLM's embedding layer to restructure representational space — maximizing the distance between semantically relevant and noisy text segments before downstream multi-label classification. Applied to a 27-label dataset of highly unstructured academic assessments, results show that while absolute F1 performance did not consistently improve, contrastive fine-tuning notably increased the correlation between mutual information and F1-scores, suggesting the method enhances LLM interpretability by making MI a more reliable predictor of classification performance in high-noise environments.

// pending DOI — IntelliSys 2026