Things I've designed, built, and kept iterating on.
A 32-bit operating system written from scratch in NASM, with a three-stage bootloader, windowed desktop, FAT12/FAT32, an RTL8139 driver and a TCP/IP stack. 17,000 lines of assembly in a 60KB kernel, and I security-audited my own kernel.
An autonomous recon and enumeration engine that plans, executes, and iterates on scanning tasks with an LLM-driven loop, wrapped in its own web dashboard.
A cyberpunk-styled node-link puzzle game for iOS. Breach nodes against the clock, choose an operator, and work through five distinct mission types.
A Python wrapper around Nmap that turns a single scan into a ready-made playbook: risk scoring, CVE/exploit lookups, and suggested follow-up commands per port.
A live tool that grades any domain's protection against email spoofing (SPF, DKIM and DMARC) from public DNS alone. No scanning, no authorisation, three seconds.
An always-on dashboard for a rotated monitor: live ADS-B flight radar, weather, bin days, countdowns and a live ISS feed. Seventeen cards, one dependency.
A fully local, voice-driven AI assistant for macOS: local LLM, local speech-to-text, a wake word, persistent memory, and a plugin system for controlling the Mac.
A self-hosted, dockerized Security Operations Center: SIEM, IDS, honeypot, and threat-intel platforms wired together to practice detection and response end to end.
An AI penetration testing platform. The agent plans and runs the test itself across 129 local tools, verifies every finding with command output as evidence rather than inferring it from a version banner, diffs each scan against the last, and writes the report.
A six-plane continuous detect-and-mitigate loop: intel triggers a probe, a confirmed finding becomes one canonical mitigation rendered per vendor, four assurance gates attack it, including an agent-generated evasion set, and a human decides before anything reaches a control.
How to build a testing harness on Anthropic’s Claude, with a methodology document generated into fourteen executable skills, scope enforced in code before the model gets control, and a human approving every command it wants to run.
A command-and-control framework written from scratch for lab work: beacons, listeners, SOCKS5 pivoting and payload staging. Transparent by design: every action attributed and logged, evasion opt-in per payload and always recorded.
A multi-agent AI pipeline that researches, writes, designs, and reviews digital products end to end, from trend research to a finished, sellable listing.