SmartSolve.jl is a Julia-based toolbox for AI-guided algorithmic discovery, designed to accelerate computations by generating enhanced algorithmic and architectural selection strategies. Envisioned as a general-purpose tool for scientific computing, current efforts focus on challenges in computational linear algebra. The toolbox addresses the growing complexity of selecting efficient solvers, data formats, precision strategies, and hardware resources for structurally diverse matrices—where conventional approaches offer substantial opportunities for improvement. SmartSolve.jl constructs a performance database through systematic benchmarking and applies automated Pareto analysis to identify optimal trade-offs between accuracy and speed. This database underpins a data-driven model that synthesizes dispatch strategies tailored to high-performance linear algebra software.

How to Cite

@software{SmartSolve2025,
  author       = {Lujan, Emmanuel and Shah, Rushil N. and Alomairy, Rabab and Edelman, Alan},
  title        = {SmartSolve.jl: AI for Algorithmic Discovery},
  month        = jul,
  year         = 2025,
  publisher    = {Zenodo},
  version      = {0.1.0-alpha},
  doi          = {10.5281/zenodo.15784217},
  url          = {https://doi.org/10.5281/zenodo.15784217},
}

Acknowledgements

We thank DARPA for supporting this work at MIT.