// portfolio — six projects, one methodology

Portfolio

Build and break with the same discipline, documented to a standard a reviewer can audit. Two living platform builds, two complete offensive/research works, applied-ML security research, and ongoing methodology-first CTF practice.

IRONVEIL Living Build

Hardened Fedora workstation built to a defence-in-depth threat model: LUKS2 full-disk encryption unlocked by a FIDO2 hardware key, remote pre-boot SSH unlock, and all DNS filtered before it leaves the host.

FIDO2 hardware-key disk unlock · 6-layer defence in depth
LUKS2Nitrokey FIDO2dracut-sshdWireGuardAdGuard Home
VIEW PROJECT →
NULLBYTE Living Build

GrapheneOS mobile platform on a Pixel 10 Pro Fold: nine independently encrypted user profiles as compartmentalised trust boundaries, verified boot from a hardware root of trust, per-app network policy.

9-profile mobile security architecture · Titan M2 verified boot
GrapheneOSTitan M2RethinkDNSTermuxWireGuard
VIEW PROJECT →
SPECTRE Complete

Grey-box penetration test of an Apache 2.4.58 host, run from a self-hardened PostgreSQL server built to CIS Level 1. PTES-aligned methodology with findings mapped to CWE, CIS, and ISO/IEC 27002.

7 findings · SHA-256 evidence chain · PTES methodology
nmapGobusterNiktoLinPEASCIS Benchmarks
VIEW PROJECT →
ORACLE Complete

Applied ML research with a security-relevant method: leakage-free image classification under class imbalance — the same discipline that governs rare-attack-class detection in intrusion detection pipelines.

93.97% from-scratch CNN vs 99.11% pretrained ResNet-18
PyTorchCNNResNet-18t-SNEscikit-learn
VIEW PROJECT →
MIRAGE Complete

Dissertation research: can frontier LLMs reason causally about social engineering, or only pattern-match its surface? Causal-discovery DAGs validated with DoWhy refutation, four frontier models benchmarked against them.

88,647-email phishing corpus · ICC 0.98 rating reliability
DoWhycausal DAGsCVAE4 frontier LLMs
VIEW PROJECT →
GAUNTLET Active

Methodology-first CTF practice on TryHackMe and HackTheBox: structured entries covering reconnaissance to root, with failure paths documented and every technique mapped to MITRE ATT&CK.

7 methodology entries · MITRE ATT&CK mapped
TryHackMeHackTheBoxMITRE ATT&CKPTES
VIEW PROJECT →

← Blog