AI researcher and engineer focused on understanding recurring AI failure modes and the underlying properties behind them. My work spans AI security, evaluation, behavioral analysis, and experimental interventions. The Samata Latch, a designed internal-state intervention for in-context binding integrity, was accepted to the NeurIPS 2026 Workshop on Foundation Models and Language Model Security (FLMSec). I am currently working on research into whether recurring AI failure invariants can inform AI architecture itself.