Avenza Security 2024 — Present
Founder & AI Security Engineer
Building a practical AI security practice for companies shipping LLM applications, agents, and AI-enabled products.
- Built RedThread — an open-source AI red-teaming and defense-evidence engine for LLM and agentic systems
- Ran adversarial campaigns using PAIR, TAP, Crescendo, GS-MCTS across 4+ attack objectives
- Implemented JudgeAgent/G-Eval scoring with OWASP LLM and MITRE ATLAS categories
- Built defense synthesis pipeline: exploit trace → scoped guardrail → replay validation → promotion evidence
- Delivered AI red-team assessments, prompt injection testing, and agentic security reviews for multiple clients