At Protiviti Global, I’m building the agentic automation layer for ITGC audit engagements. I built evidence-mapping and gap-identification agents that adapt to different clients’ control frameworks and evidence formats.
I’ve also contributed to Protiviti’s internal AI platform, working on its RAG retrieval layer and RBAC-based access-control providers. I built connectors that let agents work with the firm’s existing audit and compliance tools, and supported client security work from attack-surface mapping through verified patching.
Outside client work, I designed Fraud Analyser, a hybrid ML and LLM-agent system for payment risk detection. Its pipeline uses a calibrated gradient-boosting model for the hot lane and sends ambiguous cases to an Azure AI Foundry agent; on held-out synthetic data, it achieved 0.76 precision and 0.94 recall at a fixed 0.1% false-positive rate.
I’ve also built an open-source token-compression middleware, production multi-agent systems, and a local RAG engine with a transformer trained from scratch. My projects bring AI security into the design, including prompt-injection defenses, RBAC-scoped access, and human approval for tool use.
