Although machine learning frameworks like TensorFlow and PyTorch streamline model development, building ML pipelines that can be reliably deployed to production is still challenging. To address this problem, LinkedIn released Dagli, a JVM-based ML framework that is bug-resistant, optimizable, and accessible to both ML engineers and software engineers with less AI expertise. With Dagli, the entire model pipeline (for training and inference) is represented as a single DAG, where the root nodes are inputs to the pipeline and the child nodes are feature transformations or learned models. The framework also includes a set of pipeline components for popular models like neural networks and gradient boosted decision trees and common tasks like feature selection and model evaluation.