Most existing AI/ML techniques have the implicit assumption that the input dataset is ready in its simplistic tabular form. However, this assumption does not hold for most applications that maintain complex relational database schemas or graph structures for representing their data. In such cases, the datasets that might bring the most AI/ML insights to businesses can be scattered over and deeply buried under many highly normalized relational tables. This project has the same spirit of “Data-Centric AI” theme but from a different angle. The project’s goal is to design multi-layers of discovery and optimization strategies to bridge the gap between the two worlds (relational DBs & analytics).