Project 11 - Composable Kale Notebooks with Visual Pipeline Editor

Kale compiles a single Jupyter notebook into a Kubeflow Pipeline by parsing cell tags, detecting data dependencies with PyFlakes, and generating KFP DSL. But real ML work rarely fits in one notebook ,teams split preprocessing, training, and evaluation across separate notebooks, and today there's no way to compose them through Kale without dropping down to the KFP SDK manually. This project extends Kale with multi-notebook composition and a visual pipeline editor. Each notebook declares typed inputs and outputs (KFP artifacts like Dataset, Model, Metrics or parameter types). A new workflow compiler processes multiple notebooks and compiles them into a single Kubeflow Pipeline, with each notebook becoming a KFP sub-pipeline that preserves its internal steps. A visual editor built as a JupyterLab DocumentWidget handles assembly ,drag notebooks onto a canvas, connect typed ports, compile and run. Deliverables: #1.Notebook interface declaration system (backend + frontend) — explicit typed inputs/outputs per notebook, integrated into existing NotebookConfig and the Kale sidebar UI #2. Workflow data model and .kale-workflow file format with DAG validation (cycle detection, type checking, reference resolution) #3. Workflow compiler that processes each notebook through the existing NotebookProcessor → Compiler pipeline, wraps output as sub-pipelines, and generates a top-level KFP pipeline wiring them together through artifacts #4. CLI support: kale --workflow path/to/workflow.json #5. Visual DAG editor in JupyterLab using React Flow — drag-and-drop from file browser, typed port connections, properties panel, compile/run actions via Kale's existing RPC layer #6. End-to-end tests on a real KFP cluster, user documentation, and example workflows