AI Security · Red Teaming · Cyber
Building things at the intersection of AI and real-world risk. From red teaming to cybersecurity, I break stuff before the bad guys can.
AgentDojo is the best framework we have for evaluating prompt injection attacks on LLM agents. Its most important finding is also its most unsettling one.
Read More →Your AI Powered Accent Coach
Read More →Why function similarity matters for binary analysis, how different approaches work, and why the choice of representation is more important than people realize.
Read More →An AI accent coach that gives phoneme-level pronunciation feedback to Spanish learners — bypassing text transcription entirely to catch errors that spell-check-style pipelines miss.
A personal research portfolio built with Astro, Tailwind CSS, and MDX. An AI-native learning project — designed to be fast and easy to update and maintain