Applying Karpathy’s Loop to Build a Self-Learning Threat Hunting System

I built a continuous threat hunting model inspired by Karpathy’s auto research loop, where every hunt becomes feedback for the next one. The system ingests threat intel, builds a graph of actors, techniques, and assets, scores risk, generates hypotheses, runs hunts in Google SecOps, and learns from both positive and negative results so older threats stay in rotation when they still matter.