Gestures are an important part of our social fabric, aiding our speech (and sometimes standing in for it) in face-to-face communications. In this demonstration, we present HighFiveLive, a smartwatch application that uses data from an onboard accelerometer to detect and classify fine-grained symbolic gestures, which convey semantic meaning and have social significance. We apply logistic regression to learn a representative model of 9 symbolic gestures as performed by 32 users; in an offline evaluation study, the model accurately classifies 92% of recognized gestures. Building on this work, we deploy the learned model in our HighFiveLive mobile application, which detects and classifies symbolic gestures in real-time as they are performed. HighFiveLive is implemented as an Android application with connection to a wrist-worn accelerometer stream (such as a Microsoft Band) as well as a standalone Apple Watch application.