Abstract: Approaches are described for facilitating database queries. A data model associated with multiple data stores is received. A natural language question for a data set maintained across these data stores is obtained, where the data set is stored in a non-standardized format and not directly accessible by the querying device. An abstracted query representation is generated and translated into a standardized expression language based on the data model. Query processing approaches identify a predefined query pattern optimized for system performance metrics. An executable query is generated from the identified query pattern and provided to access the data stores. The query is executed to retrieve the data set, formatted as a structured view for unified access to dispersed data.