Solving large and sparse numerical linear systems from materials science on massively parallel supercomputers is a complex endeavor, requiring a delicate balance between accuracy and computational efficiency. Challenges include managing the scale and intricacy of these systems, optimizing scalability on parallel architectures, and addressing real-world material complexities. The research project we envisioned represents a pioneering leap in the realm of computational Materials Science, aiming to construct a cutting-edge recommender system. This recommender system is poised to revolutionize the solution of large-scale sparse linear systems, by speeding up and scaling up the linear system solutions in order to enable Materials Science research to be routinely conducted on exascale clusters. At its core, this system relies on a two-pronged approach: first, the development of a spectral predictor system, and second, the exploitation of an extensive database comprising matrices that encapsulate the essence of the surrogate space within the Materials domain. The spectral predictor system serves as the heart of the recommender, leveraging deep learning techniques to forecast spectral properties crucial for efficient linear system solving. The expansive matrix dataset, in turn, captures the diversity of spectral patterns encountered in Materials Science computations. The application of this recommender system promises to be transformative, as it will empower simulations involving hundreds of thousands of atoms, a feat previously unfeasible to achieve on current pre-exascale clusters. By linearly scaling Density Functional Theory (DFT) codes, such as BigDFT, researchers will gain the capability to simulate and analyze complex material systems with unprecedented accuracy and computational efficiency, opening new horizons for scientific exploration in the field.