Generative AI can now assemble and train a machine-learning pipeline in minutes. The hard engineering question is no longer how to build one, but how to justify that it is the right one, and to keep proving it as the pipeline evolves. When the rationale behind design choices fades, teams suffer ``decision drift'': they know what a pipeline does, but not why it remains trustworthy. In sensitive domains the stakes are real: consider an emotion-detection model for a mental-health journaling app, where users’ most intimate reflections must be anonymized and protected. This tutorial uses jPipe, a textual DSL, to model the argumentation behind such pipelines as lightweight justification models and to make those models executable. Participants will learn to (1) capture engineering rationale as justification models traceable to requirements, (2) compose justifications across stakeholders (data scientist, security engineer) and instantiate reusable patterns, and (3) execute validation guardrails that run automatically in CI/CD pipelines. By the end, attendees turn argumentation from documentation into continuously enforced conformance checks. No machine-learning background is required.