breandan

Type-safe automatic differentiation framework enabling users to express differentiable programs with higher-dimensional data structures and operators. Ensures compile-time algebraic validity, reducing runtime errors and supporting advanced features like shape-safe tensor operations, symbolic derivatives, and property-based testing for numerical gradient checking.

Implements a novel computational model for graph computation, translating to iterated matrix multiplication on GPUs. Supports algebraic circuits, neural networks, proof networks, and various propagation schemes using message passing. Provides visualization, translation between graph formats, and tools for regex to NFA compilation.