Matthew Gresham

AP Courses: Computer Science Principles (2025–26, A+), English Language & Composition (2025–26), Computer Science A (2026–27, in progress — Java), Cybersecurity (2026–27, in progress)
Honors Courses: HS Band (2025–27, in progress), Tech Theatre (2025–27, in progress)
Other Coursework: Pre-Calculus (2025–26), Physics (2025–26), Philosophy & Worldview (2025–26)
Additional: Codecademy Intro to IT, BYU United States History

• Lead a three-person testing team, owning the QA process end to end for an all-in-one academic platform covering grade tracking, an AI study assistant, college planning, and FAFSA deadline tracking.
• Run manual test passes across the platform, file and triage bugs before releases, and write automated checks to catch regressions on core flows.
• Maintain a running testing log of bugs, UX friction points, and feature suggestions found across signup, the AI assistant, calculator tools, and account settings; mentor team members through specific tech topics and platform components.

• Maintain and modernize the front end of an existing neuroscience education site, turning designs into responsive, accessible pages and keeping the web presence current as the project grows.

• Led the buildout of a Bangkok-based non-profit’s social-media presence across Line, Facebook, and Instagram, from zero prior organizational presence.
• Advised the foundation on content strategy and posting cadence as the project’s sole technical/communications lead.

• Conceptualized, architected, and developed an automated security-evaluation tool that detects and mitigates prompt-injection and adversarial attacks against Large Language Models (LLMs).
• Built an attack-prompt library spanning prompt injection, DAN jailbreak, roleplay bypass, and system-prompt extraction, plus a FastAPI backend that fires them at local Ollama models and streams pass/fail results to a React/Vite frontend as NDJSON.
• Added a headless benchmark pipeline that grades and compares multiple models with a severity-weighted risk score, with a pytest suite and CI, and full WSL2 support.
• Mapped identified vulnerabilities against emerging AI-security frameworks (OWASP LLM Top 10) to ensure robust model defenses and secure deployment.

• Designed and shipped a deployed full-stack platform that AI-matches learners into study cohorts, with real-time discussion, AI facilitation, and a Study Hub offering Socratic tutoring, essay coaching, and study planning.
• React + Vite + Tailwind frontend and a FastAPI backend, both live on Vercel (serverless Python), backed by Supabase (PostgreSQL + auth + realtime), at zero cost.
• Wired a GitHub Actions CI gate on pull requests and ran free-tier OpenRouter LLMs to keep inference costs at zero.

• Full-stack financial tool featuring both a command-line interface and a multi-page Streamlit web application.
• Includes currency conversion, budget tracking, tax calculations, and interactive data visualizations.
• Handles session state management, field validation, and persistent data storage across user sessions.

• Built a self-hosted vulnerable-web-app server with a custom bash setup to practice SQL injection, cross-site scripting (XSS), and brute-force attacks in a controlled environment.

• Magellan Search Engine — Go web spider + React frontend deployed on Vercel; indexes content into SQLite
• Markdown to HTML Converter — Go CLI tool that converts Markdown files into styled HTML pages
• Chess AI — Python chess engine using minimax with alpha-beta pruning, expanding toward a self-play Deep Q-Network

• Built and submitted a privacy-first RAG study tool that ingests Markdown files, PDFs, and images into ChromaDB for conversational querying.
• Runs fully offline via local Ollama inference so no data leaves the machine, with cloud models (Groq, ChatGPT, Anthropic) as an option for low-end hardware.
• Shipped a setup.sh script for one-command installation on Linux, macOS, and Windows, plus an auto-update feature.
• Dual interface: CLI for lightweight use and a Streamlit Web UI with live camera upload for photographing physical notes.

• Built and deployed a 6-agent CrewAI pipeline that accepts any data file and a plain-English question, then routes the work through a sequence of specialist agents — context rewriter, data cleaner, prompt engineer, analyst, output formatter, and QA critic. Solo entry.

• Designed, built, and submitted Villages (featured in Technical Projects above) under hackathon time constraints, shipping a deployed full-stack AI learning platform end to end.