While machine learning shows promise in automating decision-making, human involvement remains crucial in the complex machine learning lifecycle, from data collection and preparation to model development and deployment. Current lifecycle management tools often expose low-level primitives that misalign with users' intentions to specify high-level semantics and objectives. This misalignment makes it challenging to effectively incorporate human insights into data analytics, requiring significant manual efforts or leading to ad-hoc heuristics that compromise model accuracy and system efficiency. In this talk, I will present our lab's initial steps in developing a human-centric toolkit for machine learning lifecycle management. We will focus on interfaces and programming models that streamline user involvement in data preparation, labeling, and cluster management.