Ian de Marcellus

Security researcher & software engineer, focused on AI systems

About

I'm interested in security, AI, and understanding complex systems. Currently focused on LLM internals, fine-tuning, and alignment: both the technical foundations and the harder questions about how to build these systems well. Based in San Francisco.

Harvey Mudd College — BS with High Distinction, Joint Mathematics & Computer Science

Selected Work

Security Vulnerability Research

Discovered and responsibly disclosed two security vulnerabilities to Apple with proof-of-concept demonstrations.

BGP Security Research

Conducted analysis of BGP protocol vulnerabilities and demonstrated hijacking attacks in a controlled five-router environment. Academic clinic project with Juniper Networks.

Incident Timeline MCP Server

Open-source tool for extracting structured data from incident response logs. Built with Python and the Model Context Protocol.

LLM Fine-tuning Research

Ongoing experiments with LoRA fine-tuning, custom training data, and alignment approaches. Exploring how training methods shape model behavior.

Links