Scrapsort is a mechatronic system designed to sort recyclables at the edge. It aims to provide an efficient and inexpensive solution to the growing problem of waste. At the heart of Scrapsort lies an ultra-low-power microcontroller, the MAX78000, which includes a hardware accelerator for convolutional neural networks (CNNs). Scrapsort implements a lean machine learning model that can identify and categorize various recyclable materials, such as glass bottles and aluminum cans, while in motion on a conveyor belt. As the items travel down the conveyor, a series of pneumatic cylinders sort the recyclables into storage bins. Simultaneously, a robotic arm leverages a pneumatic suction cup to separate more valuable recyclables from the process stream. Scrapsort is designed to frontload the task of categorizing waste, so that it can be processed more efficiently and salvaged to the fullest extent possible.