Flywheel
An autonomous, multi-agent platform that researches software for security vulnerabilities and verifies them before reporting.
Vulnerability research is slow, manual work, and AI tools tend to invent “findings” that don't actually reproduce — wasting triage time and eroding trust.
I designed and built the whole system: a staged decision graph that maps a target, forms a hypothesis, drafts a proof-of-concept, and only reports a finding once a deterministic validator reproduces it end-to-end.
Reproduced proof-backed findings across 20+ open-source projects under a strict reproduce-before-report standard. Built strictly for authorized testing.