Focused on the security of LLMs and agentic systems through adversarial testing and vulnerability research.
I work on prompt injection, agent/tool interactions, RAG security, MCP security, authorization boundaries, and emerging attack surfaces in AI-powered applications.
Previously focused on blockchain security, where I developed a strong adversarial mindset through audits, vulnerability research, and protocol analysis.
Python · Rust · Go · C · JavaScript
AI & Agent Security
Red-team LLM and agentic systems across indirect prompt injection, RAG, MCP/tool trust boundaries, authorization failures, and agent-driven attack chains — with an emphasis on real-world exploitability and impact.
Vulnerability Research
Discover and validate vulnerabilities in AI-powered applications, trace them to their root cause, and build reproducible attack chains and PoCs covering access control, persistence, tool abuse, and data exposure.
────────────




