devival

I build autonomous AI agents and evaluate their security capabilities — then audit the blockchain protocols they interact with.

As a Tier-1 smart contract auditor (Uniswap V4, Reserve Protocol, ABDK Consulting, Oak Security), I have found and reported vulnerabilities in production code securing $1B+ in user funds across 100+ audits. As an AI engineer, I design and measure multi-step LLM agent systems: discovery → adversarial verification → judge-layer scoring against ground-truth vulnerability corpora.

The intersection — AI applied to adversarial security — is where I focus. I built vuln-agent-eval, an open dangerous-capability eval harness measuring how well autonomous LLM agents find real smart-contract vulnerabilities. I hold 8 Anthropic certifications across the full Claude developer stack: Claude API, Claude Code, MCP, Agent Skills, Vertex AI.

Open to: AI security research · LLM agent engineering · frontier red-team / eval work · technical advisory at AI-focused VCs and PE firms.