I am an independent AI safety researcher and systems engineer building open-source safety infrastructure under real-world constraints. My work focuses on offensive safety methods — adversarial red-teaming, automated stress testing, and reproducible evaluation — as tools for prevention rather than exploitation.

I am the creator of Red Set ProtoCell, an open-source framework designed as an “immune system for AI,” using a dual-agent Sniper and Spotter architecture to proactively surface, classify, and log model failure modes. I design and deploy full-stack systems with an emphasis on transparency, auditability, and practical usability for independent evaluators.

My motivation for this work comes from a long-standing focus on accountability and harm reduction in high-impact systems. I am particularly interested in safety tools that empower smaller researchers, external auditors, and the public to meaningfully evaluate AI systems, rather than relying solely on closed, internal assurances.