MLOps Continuous Training System
Key developer in the Kubernetes refactoring and modernization of a multi-tiered triage pipeline, typical of large organizations maintaining cyber defense. The system processed a continuous stream of artifacts—executables, scripts, and observables—that required automated enrichment, classification, and prioritization.
Primary contribution was the design and implementation of a self-managed Zeebe broker BPMN-based orchestration layer that replaced ad hoc scripts with structured, declarative workflows. I built support for defining analytic pipelines as composable DAGs, integrating tools such as YARA, Ghidra, and internal classifiers as reusable analytic tasks.
This work focused on a BPMN-based orchestration system designed to streamline triage operations, improve maintainability, and support future scaling:
Developing the workflow execution engine and task interface
Managing integration with existing enrichment tools and data sources
Supporting task routing and branching based on artifact scoring or metadata