Validated Computation
AI-guided Computer Algebra System (CAS) computation and validation are implemented and producing results across a broad range of mathematical problems.
Live Development Status, May 2026
The architecture is implemented and running; results are being actively measured.
ExaktAI tackles the trust problem in AI mathematics by building a validation architecture around it.
AI-guided Computer Algebra System (CAS) computation and validation are implemented and producing results across a broad range of mathematical problems.
ExaktAI operates with six independent AI systems: Claude, Codex, Gemini, DeepSeek, Grok, and Mistral. Validation draws on their combined output and the CAS computations they drive, not on trust in any single system, AI or CAS.
The desktop App accepts natural-language problems, native 2D mathematical notation, and direct CAS input. No particular input mode is required.
ExaktAI runs from the ExaktAI App and directly from the Maple user interface. The validation architecture is independent of the interface used to reach it.
Results are delivered as Mathematica notebooks or Maple documents: executable, auditable, editable, and shareable by you, on the platforms the scientific community already uses. Validation status is embedded in the document itself.
A 100-problem benchmark covering 14 areas has been run across all AI systems. Analysis of these results is in progress and driving improvements to the validation architecture.
ExaktAI already uses Mathematica for validation and produces auditable Mathematica notebooks. It can also be launched from within a Maple document. Launching it from within a Mathematica notebook is in progress.
ExaktAI currently runs in a controlled development environment. Making it accessible to external users (researchers, educators, and collaborators) is in preparation.
ExaktAI currently handles the undergraduate mathematics core well; completing the mathematics core and adding undergraduate physics is in progress.
Using the ISED / Innovative Solutions Canada TRL scale:
ExaktAI matches TRL 6: a prototype in near-desired configuration, tested against a multi-hundred-problem benchmark in a controlled environment. TRL 7 requires the system to be accessible to users outside the development environment, which is the Public Access milestone currently in progress.