While training a model is usually a batch function done in safe private areas, scoring data as it is received is a real time event. When it comes to ETL and data ingestion pipelines real-time up-to-date results with a low memory footprint is a must. To build a reliable, performant, at-scale data acquisition system you must have the right streaming/messaging platform. In this session we will look at how Apache Pulsar, a state-of-the-art messaging system, enables construction of powerful and articulated data processing/ingestion pipelines.