In this presentation, we will talk about what we have learned from our experience building a horizontally scalable deep learning pipeline that can process the > 120 million images in the OpenStreetCam database. We will offer a technical perspective about how we used state of the art algorithms to detect a large number of traffic signs, cluster those detections together and finally extract the real-life GPS locations of the detected signs. This technology is now open source and available for anyone to use and improve upon