Embodiments store attributes extracted from incoming media data (e.g., image, audio, video), in a media store residing in a data lake together with other, non-media attributes. In response to incoming media data (e.g., an image), an engine references an unpopulated media attribute schema resulting from processing a trained deep learning model (e.g., a Convolutional Neural Network—CNN model). The engine applies the deep learning model to extract from the incoming media data, a media attribute (e.g., a cloudy spot dimension) comprising a prediction value and a confidence. The engine populates the media attribute schema with the attribute (value; confidence) and an identifier, and stores the populated media attribute schema in the data lake. The data lake also includes a non-media attribute (e.g., patient info) sharing the identifier. Now, the data lake may be queried for both the non-media (patient info) attribute and the media (image) attribute extracted by the model.