~/jw3 — zsh — 96×24

$whoami --verbose

james wilson // iii

cybersecurity intern @ NYC School Construction Authority · M.S. cybersecurity, Fordham '26

location

Brooklyn, NY

status

open to full-time roles

focus

adversarial ML · network defense · agent security

last_updated

2026-05-13

$# scroll for project index, writing, and resume

about

01 // README.md

I came up through Computer Science & Mathematics at St. Joseph's University in Brooklyn — the math half is still where most of my best instincts come from.

Graduate work at Fordham pulled me toward adversarial ML, secure AI, and threat hunting. Most of what I build now lives at that seam: detecting attacks, hardening pipelines, and writing the audit trail that lets a system answer for itself.

Off-hours I tinker with my home-lab, run Linux a little too aggressively, watch a lot of film, play Pokemon Champions, and walk through my borough on weekends.

// currently

readingCySA+ Material
listeningIT'S BEEN AWFUL by Isaiah Rashad
buildingthis website!
watchingVinland Saga S2
distroArch (btw)

$ ls ~/scrapbook/ --recent

James in Arashiyama, Kyoto, summer 2024 — // arashiyama, kyoto — st.joseph's travel-study, '24 JPG

James on the subway home from NY Comic Con — // q train, post-NYCC. peak brooklyn. JPG

work

02 // projects/

001 research

Adversarial Detection for Video-Based Face Recognition

Dual-detector framework — temporal drift + ensemble disagreement — catches ATN, PGD & FGSM attacks on FR systems with 89.3% accuracy.

pytorchu-netsklearn

↓ paper →

002 prototype

Secure AI Agent Middleware

Runtime security layer for agentic LLMs — normalization, policy, contagion detection. Drops attack success rate to 0% across a 12-vector suite.

pythonllm-sechtml5lib

↓ paper →

003 complete

AI-Phishing Detection via Communication Drift

Per-sender stylometric profiles + DriftScore flag impersonation emails that header and URL filters can't see.

sklearnstylometrynlp

↓ paper →

+ next project
// coming soon

writing

03 // posts/

2026.04.18 Why temporal drift is the cheapest signal in video face-recognition defense 9 minadv-ml 2026.02.03 Communication drift — catching AI-generated phishing without URL filters 12 minnlp 2025.11.21 Agent security is a system problem, not a prompt problem 7 minllm-sec 2025.09.06 On building a home-lab that survives my own changes 5 mininfra

resume

04 // experience.log

2025 → present

NYC School Construction Authority

cybersecurity intern

Vulnerability triage, internal tooling, and policy work for a network supporting every public-school construction project in the city. Lots of practical incident-response and threat-hunting reps.

2023 — 2025

Fordham University

M.S., cybersecurity

Graduate work in adversarial ML, applied cryptography, and network defense. Thesis-adjacent research on face-recognition robustness.

2019 — 2023

St. Joseph's University, Brooklyn

B.S., computer science & mathematics

Dual major foundation in algorithms, probability, and systems. The math discipline still shows up in everything I do.

contact

05 // reachable.json

say hi.

// open to full-time security & ML-engineering roles starting summer 2026