I'm Paul Harrison.
I build software. Lately, it talks back.

Computer Science @ Michigan | AI + SWE + Security

I design and ship full-stack products across modern web stacks and cloud infrastructure. My recent work spans voice agents, LLM pipelines, Vue + TypeScript frontends, FastAPI services, and agentic systems — built with a focus on reliability, performance, and strong engineering fundamentals.

View Projects Experience Resume

About Me

Get to know me

I'm Paul Harrison, a Software Engineering-focused Computer Science student at the University of Michigan. I enjoy building production-minded systems where clean architecture, strong UX, and measurable impact all matter.

From backend APIs and data workflows to frontend product experiences, I like owning features end-to-end. I care about pragmatic engineering: writing maintainable code, validating with real users, and continuously improving performance and reliability.
Feel free to contact me here.

My Skills

Ollama

LiveKit

Whisper

Hugging Face

LoRA

LiteLLM

Venice AI

LLM Evaluation

Schema-constrained JSON

Frontend

React

TypeScript

JavaScript

Backend

Python

C++

FastAPI

Flask

PostgreSQL

Firebase

SQL

Infra

Docker

Google Cloud Run

Git

Web Scraping

Technical Experience Internship experience building and shipping AI-driven software in startup and product environments.

Whomlab · Full-Stack Software Engineer Intern

Prague, Czech Republic | Jan 2026 - Present

Built a voice-driven survey platform projected to serve 7,500+ users across five EU-financed programs. Engineered a LiveKit WebRTC pipeline (browser mic → LiveKit Cloud SFU → Python livekit-agents worker with Whisper STT → Venice AI LLM → OpenAI TTS). Shipped Vue 3 + TypeScript frontend with a custom useLiveKit composable, MediaRecorder capture, data-channel dispatch, and Pinia state — backed by FastAPI + Firebase Auth, Firestore, and Cloud Storage.

Charlemagne Labs · AI & Security Software Engineering Intern

Ann Arbor, MI | Aug 2025 - Dec 2025

Fine-tuned Gemma 3 (270M) with LoRA adapters for phishing URL classification, merging weights with a custom classifier head for dependency-free inference in a FastAPI serving layer. Built a multi-backend evaluation harness (Hugging Face, Ollama, LiteLLM for GPT/Gemini) benchmarking accuracy, precision/recall, and latency across realistic phishing and social-engineering scenarios. Engineered URL, DOM, and headless-browser feature extractors — domain signals, password-field and urgency-language detection, image hashing for look-alike site detection.

SupermaxCloud · Cybersecurity & Infrastructure Engineering Intern

Chicago, IL | May 2025 - Aug 2025

Automated vulnerability scanning across internal and internet-facing systems with Nmap and Shodan, cutting manual review time by 70%. Managed infrastructure security via firewall rules, access controls, and Wazuh-based monitoring. Built Zabbix and Kuma dashboards and health checks, improving visibility into uptime and service issues across the fleet.

Projects Selected software projects with a focus on full-stack engineering, data systems, and production deployment.

Whaleden Comic Artist Discovery System

An agentic data pipeline that discovers and ranks 1,400+ comic artists across Bluesky, Reddit, and Dribbble. Features a 4-stage Ollama (Llama 3.1 8B) LLM pipeline with schema-constrained JSON outputs and confidence thresholds, alongside a Streamlit browsing UI, resilient scraping, deduplication, and normalized Postgres storage.

Simon Smiles

Real-time facial expression game powered by Flask, OpenCV, and FER inference, with a responsive browser UI and webcam integration. Deployed to Google Cloud Run with production HTTPS and CORS configuration.

OR CHECK ME OUT!

PassSec

Full-stack web app for Security+ preparation with React + Tailwind frontend, Flask REST API, JWT authentication, and PostgreSQL-backed quiz content. Features domain-based filtering, history tracking, and a deployment plan for Google Cloud Run and Netlify.

OR CHECK ME OUT!

Bonus Projects, Labs, & Certs Hands-on labs, mini-projects, and exercises exploring cybersecurity concepts, risk assessments, and security tooling.

SecVault

End-to-end encryption directly in the browser with AES-256 applied client-side before upload, so private files remain private from source to storage.

CHECK ME OUT!

Google Cybersecurity Professional Certificate: Mini-projects, labs, notes

Comprehensive notes, detailed lab write-ups, and mini-projects covering risk management, network security, system hardening, SIEM analysis, incident response, and digital forensics.

Explore!

Passive Recon – [REDACTED]

Performed passive reconnaissance on a live domain using DNS, WHOIS, Shodan, and DNSDumpster. Created a detailed PDF report including findings and security recommendations. Attached is a redacted version.

View Full Report (PDF)

Anonymized Home Network Topology

A scan & diagram of my home LAN (192.168.1.0/24) created with Nmap and manually anonymized for this portfolio. Highlights device roles, IPs, and network flow from WAN→Firewall→Router→Switch→Clients.

View Full Diagram

I'm Paul Harrison.I build software. Lately, it talks back.