Matt G

About Me

Matt Gresham


I'm a self-taught developer building toward a career at the intersection of AI/ML, cybersecurity, and software engineering. My stack is Python for logic-heavy work and Go for performance-critical tools, with React when the project needs a frontend.

I started programming to scratch my own itch — the Finance Kit began as a school submission and grew into a full-stack deployed app. That pattern repeats across my work: I build something I'd actually use, then push it until it's production-ready.

Current focus: cybersecurity tooling, machine learning fundamentals, QA work as Head of Testing at Polaris Student, and web development for Neurole.org.

View Projects View Résumé Contact Me

Looking for the academic record — coursework, awards, activities, service, and college plans? That lives on the Academics page.

Work Experience


Polaris Student — Head of Testing

Head of Testing for Polaris Student (polarisstudent.com), an all-in-one academic platform for high school students (grade tracking, AI-powered study help via its Polly AI assistant, college planning, and FAFSA deadline tracking). Lead a three-person testing team (myself included), mentoring the other members through specific tech topics and platform components as needed. Own the QA process end to end — run manual test passes across the platform, file and triage bugs before releases, and write automated checks to catch regressions on core flows. Keep a running testing log of bugs, UX friction points, and feature suggestions found across signup, the AI assistant, calculator tools, and account settings.

View Testing Log

Neurole — Web Developer

Web developer for Neurole (neurole.org), maintaining and modernising the site's front end. Turn designs into responsive, accessible pages and keep the web presence current as the project grows.

Saphan Siam Foundation — Social Media Lead (Eagle Scout Project)

Led the effort to build a Bangkok-based non-profit's entire social-media presence from scratch as my Eagle Scout service project — sole technical and communications lead across Line, Facebook, and Instagram. Created content plans, set up platform profiles, and advised on how to reach a wider community audience through each platform's native features and posting cadence.


I also take on freelance and client work — automation, dashboards, web apps, AI integrations, and security/QA.

See what I build for others →

Flagship Projects

My strongest work — each deployed, tested, and documented in detail.

LLM Protector Used by real people

Security scanner probing local LLMs for prompt-injection/jailbreak vulnerabilities — 50 attacks across 11 categories, severity-weighted risk scoring. Live demo · Project page

Stock Analysis Engine Used by real people

Full-stack finance tool with live data, technical indicators, and an AI analyst grounded in computed metrics. Shared core powers both web (FastAPI+React) and offline (Streamlit) modes. Live demo · Project page

The Finance Kit Used by real people

Deployed finance tracker (Python + Streamlit) with expenses, income, subscriptions, budgets, tax estimates. Clean FE/BE split with shared data layer. Live demo · Project page

Magellan Search Engine Used by real people

From-scratch search stack: Go web spider + SQLite index + React frontend. Built to understand how indexing and ranking actually work. Live demo · Project page

All 30+ projects →


How I Work


Build to learn, not to demo

Every project starts with a problem I actually have. The Finance Kit began as a school submission; LLM Protector came from needing to test local models. I push until it's production-ready — CI, tests, deployed, documented — because that's where the real learning happens.

Offline-first, security-minded

tinyGPT, MedicalAI, the Finance Kit's offline mode — I build for environments without connectivity or API keys. Security isn't an afterthought: LLM Protector red-teams models; Petri sandboxes untrusted code; the Cyberdeck is air-gapped by default.

Self-taught ≠ solo

I learn by building, but I ship with others. Polaris Student: leading a 3-person QA team across 13 modules, 8 languages. Neurole: modernizing a live product with a CSV-driven content pipeline. Hackathons: solo and team entries, all deployed. Client work: from ARGs to Stripe+Coinbase billing systems.

Show the work, not the resume

This site is the resume. Every skill listed on the Projects page is counted from real project cards — click a skill chip to see the evidence. No self-assessment bars, no "proficient in X" without a repo to prove it.


Hackathon Highlights


International AI Agents Hackathon 2026

6-agent CrewAI pipeline (context rewriter → data cleaner → prompt engineer → analyst → output formatter → QA critic). Frontend Vercel, backend HuggingFace Spaces. Solo entry. View submission

Hack America 2026

Study Assistant — RAG tool ingesting Markdown, PDFs, images into ChromaDB for conversational querying. CLI + Streamlit UI. Originally prototyped as LLM Protector's study module; pivoted to standalone tool due to hackathon time constraints. View submission

USAII 2026 — Villages

AI learning platform: cohorts with AI matching, real-time discussion, AI Village Elder, Study Hub with Socratic tutoring. FastAPI + React + Supabase. View submission

HackMars 3.0 — Calendar AI Assistant

4-agent CrewAI pipeline connecting to Google Calendar/Gmail via OAuth. Natural language schedule management. Teammate: Mackenzie. View submission

All 6 submissions →


Currently Working On

Three active builds in my AI Lab: tinyGPT, a character-level GPT written from scratch in PyTorch that trains and runs on any CPU (built to be an offline lab assistant on a plane); note: tinyGPT was the predecessor architecture that informed Clyde's design and is no longer actively developed. MedicalAI — Light Weight, a research project fusing CLIP chest X-ray embeddings with Bio_ClinicalBERT symptom text into a lightweight, offline diagnosis model for rural care; and Clyde (GitHub), an AI bot in training with persistent memory, tool use, and a modular skill system. Current focus: long-term memory consolidation, multi-step reasoning, and safe autonomous operation.

Also submitted Villages to the USAII Hackathon, creating study programs to study for finals, and working on selling a SaaS application legally.

Alongside this I'm working through ML fundamentals (gradient descent, backprop) as I build toward more applied AI/ML work.

What I’m Learning Now


I learn a new tool by building something real with it. Each skill below is paired with the project I’m learning it on, so you can see how far along it is.


Résumé


View my résumé — kept current by hand; use your browser's Print to PDF to save a copy.

Back to top