GPS incurs large errors (often 50 meters or more) in urban environments due to line-of-sight blockage from buildings, and interpretation of strong reflections as line-of-sight. In this talk, we describe ShadowMaps, a Bayesian framework for significantly improving urban localization performance by probabilistic shadow matching,  exploiting information on satellite signal strengths (available from GPS receivers via a software interface in Android devices) together with 3D maps. The framework also enables crowdsourced creation of 3D maps via Shadow SLAM. a Simultaneous Localization and Mapping technique tailored to the information obtained from shadow matching. We also sketch the technology transfer story of how ShadowMaps is now being integrated into Uber’s location infrastructure.