Calcium sulfoaluminate cements (CSACs)—for which CO2 emissions are ~50% lower compared to Portland cement (PC)—present a tremendous opportunity to develop sustainable binders for construction infrastructure. To further reduce the energy-intensity and carbon footprint of CSAC, supplementary cementitious materials (SCMs: e.g., a mixture of limestone and fly ash) can be used—at least in theory to replace up to 50% of the CSAC in the binder. That said, owing to the substantial diversity in SCMs’ compositions—plus the massive combinatorial spaces, and complex SCM-CSAC interactions—current computational models are unable to produce a priori predictions of properties of [CSAC + SCM] binders. This study presents a deep learning (DL) model capable of producing a priori, high-fidelity predictions of composition- and time-dependent hydration kinetics, phase assemblage development, and compressive strength development in [CSAC + SCM] pastes. The DL is coupled with a thermodynamic model that constrains and guides the DL, thus ensuring that predictions do not violate fundamental materials laws. The training and outcomes of the DL are ultimately leveraged to develop a high-fidelity prediction tool to determine optimum precursor chemistry and mixture designs of [CSAC + SCM] binders that exhibit superior compliance-relevant properties compared to PC concretes, while restricting the CSAC content to ~50%.