Show HN: Chelis, a tensor language agents can write and prove (github.com)

Chelis has emerged as a groundbreaking numerical computing language specifically designed for artificial intelligence agents to write, validate, and run code with human oversight. With features such as named tensor dimensions, precision type checks, and a comprehensive proof stack, Chelis enables agents to rigorously check properties of the code before execution. For instance, the language handles operations like weighted summation by ensuring tensor rank and dimension compatibility, significantly reducing runtime errors caused by mismatched tensor shapes that are common in languages like NumPy. This development holds substantial significance for the AI/ML community as it streamlines the coding process with enhanced safety checks and validation mechanisms, fostering trust in automated systems. Chelis includes innovative tools like chelisup for easy installation of toolchains and functions to cross-check code quality and validity against predefined properties. The language not only supports general-purpose numerical computations but also facilitates the transition of code to C for broader application. By mechanizing certain aspects in Lean 4 and providing accessible error reporting in JSON format, Chelis represents a leap towards safer and more reliable AI programming practices, encouraging higher quality and robustness in machine learning applications.