This paper attempts to situate concrete recycling within the contemporary economically convenient waste streams. In the case of concrete, the economically convenient waste stream is to demolish it into rubble and take it to landfill or grind it into aggregates. This research introduces a novel method using machine learning algorithms to analyze and match concrete irregular rubble surfaces, transforming them into a new assembly called "Reused Rubble Concrete Masonry Units" (RR-CMU). The process involves digitally scanning rubble, fixed length vector candidate extraction and a machine learning assisted matching process capable of searching for and aligning data-rich digital meshes. The output of this matching process is a modular structural unit produced within a factory quality control setting. Case studies demonstrate the performance of the matching process for both simulated data sets and a digitally scanned set of real-world rubble. We present results to demonstrate the quality of the matching with one-to-one matched rubble examples. This paper shows the viability of this new upcycling workflow for the construction of RR-CMU directly from a waste stream and demonstrates a ninety percent reduction in embodied carbon (kgCO2eq) when compared to conventional concrete construction.