To build effective machine learning systems, model developers must understand how a model might perform on different datasets. As such, data practitioners must build an intuition for how different types of model architectures perform on various datasets (e.g. with different distributions). Drawdata helps data practitioners cultivate this skill by enabling users to generate their own datasets by drawing lines or scatter charts (which visually represent the underlying data). With the datadata Python package, users can draw a dataset in a Jupyter notebook, copy it to their clipboard, and then read it in with pandas.