It started at 9, guessing the passwords on my parents' devices so I could play games. Still at it: 18 now, studying IT at QUT and hunting on HackerOne. Most of what I ship is security and AI tooling - an autonomous bug-bounty agent, an authorization-first LLM red-team framework, a Burp-to-MCP bridge, a hardware-agnostic LLM inference engine, and a browser runtime built for AI agents.
offensive security · applied ml · agent systems · systems programming
|
Autonomous bug bounty hunting agent. 71K lines of Python. Hypothesis-driven pentesting with 19 reasoning modules, 51+ tools, and HackerOne integration. |
Authorization-first LLM red-team framework. Local 70B abliterated attacker LM, 6 novel transformer-architecture attacks, statistically-validated universal jailbreak claims. |
|
Burp Suite extension that exposes the Montoya API as an MCP server. 152 typed tools + cross-extension reflection bridge into Logger++, Hackvertor, Param Miner. Drives Burp from any AI agent. |
Browser runtime built for AI agents. Semantic tools, persistent site memory, self-healing execution, and a built-in MCP server. |
|
Hardware-agnostic LLM inference engine — its Sparse Oracle Architecture runs anything from a Raspberry Pi to a B200 cluster. Written in Rust, shipped on PyPI.
|
Local-first reverse engineering orchestrator. Drives Ghidra, angr, Frida, rizin, QEMU, AFL++ and pwntools through a small-model LLM to triage binaries and find bugs. No API keys, no network calls. |
offense
ai / ml
systems
ops
- `push` [AshtonVaughan/Nerve](https://github.com/AshtonVaughan/Nerve) - docs: mark project as non-working and unmaintained
- `push` [AshtonVaughan/AnyData](https://github.com/AshtonVaughan/AnyData) - docs: mark project as work in progress
responsible disclosure only / AEST / always learning