I built FraudShield, a reproducible fraud-risk pipeline with versioned stages, a typed inference API, and a regression-aware promotion gate. On held-out synthetic data, it measured 0.838 ROC-AUC and 0.757 recall, with PSI drift reporting and Prometheus alerts.
I also built RAGOps, a source-grounded RAG service with citations, safety guardrails, and release gating, and ModelGuard, a fail-closed model-admission gate that generates AI-BOM and SARIF evidence. My project work includes automated tests, infrastructure as code, and observable operations; I also developed React, Node.js, Express.js, and MongoDB applications at Tooroot Consulting Services.
