Résumé : Although we have reached new levels in smart city installations and systems, efforts so far have focused on providing diverse sources of data to smart city services consumers while neglecting to provide ways to simplify making good use of them. In this context, one first step that will bring added value to smart cities is knowledge creation in smart cities through anomaly detection and data annotation, supported in both an automated and a crowdsourced manner. We present here LearningCity, our solution that has been validated over an existing smart city deployment in Santander, and the OrganiCity experimentation-as-a-service ecosystem. We discuss key challenges along with characteristic use cases, and report on our design and implementation, together with some preliminary results derived from combining large smart city datasets with machine learning.