I'm Will, an Adversarial Machine Learning researcher and Agentic Systems Engineer operating at the boundary of ML security and next-generation AI infrastructure. My work targets the full AML attack surface — from gradient-based model perturbations to ecosystem-level exploitation of autonomous agent orchestration — then engineers the hardened infrastructure that survives it.
At Noctis Research, I design and execute AML attack campaigns against production AI deployments. This spans evasion attacks on deployed classifiers and safety layers, poisoning of RAG corpora and fine-tune pipelines, model extraction and inversion against API endpoints, and full agentic ecosystem exploitation: privilege escalation via tool chaining, goal hijacking through indirect prompt injection, and persistent state corruption in multi-agent memory.
At Zyplabs, I engineer the defensive side: adversarially-hardened agentic AI infrastructure templates — RAG pipelines built to resist poisoning, microservices hardened against extraction, and agentic orchestration with anomaly-monitored tool-use surfaces.
Technically, I operate in Python-native AI/ML stacks with deep roots in Linux systems engineering, FreeBSD and NixOS hardening, DevOps pipelines, and production RAG architecture.
