Deploying AI in healthcare requires more than accurate models—it demands strict reproducibility, traceability, and compliance. In this talk, we explore how Bazel enabled a fully reproducible, audit-ready ML pipeline used in clinical radiology. We demonstrate how Bazel was used to manage Python, C++, and Docker-based components, enforce hermetic builds, lock dependencies, and generate Software Bills of Materials (SBOMs) for compliance. With Bazel, we achieved deterministic model training, sandboxed preprocessing, and secure inference packaging across multiple hospital environments. Attendees will gain insights into structuring Bazel for regulated machine learning workflows, managing multi-language codebases, and building trust in sensitive AI systems. This session offers practical strategies for engineering reproducible, scalable infrastructure in real-world clinical settings.