Local atomic environment (LAE) representations have enabled a deeper understanding of complex alloy configurations, by enabling machine learning and data science approaches to previously inaccessible tasks. For example, recent developments have used site-centered descriptors to interpret the structurally complex LAEs present in polycrystalline grain boundary (GB) networks and learn the spectrum of grain boundary segregation energies available across many binary alloys. One limitation of many LAE representations is their “site-focus”, which limits their ability to capture chemical complexity at the atomic scale. This paper introduces a modified “bond-focused” LAE representation, which is used to assess pair-wise solute interactions in the topologically complex environments of grain boundaries. This approach opens the pathway to learning models, which we develop here to construct a large-scale database of spectral parameters describing grain boundaries in binary alloys.