In my spare time, I built Noema, a declarative AI programming approach that makes it easy to interleave Python and LLM (Large Language Model) generations.

GitHub Repository: https://github.com/AlbanPerli/Noema-Declarative-AI

In a nutshell: Noema enables prompt programming using Python.

The longer story: For humans: Noema generates a kind of ReAct prompting that interfaces seamlessly with Python. It allows you to express algorithms as a combination of semantic reasoning (from the LLM) and classical algorithmic logic. The goal is to provide a new way to think about programming—merging traditional code with the flexibility and power of AI. (Whether this approach proves practical or not is still an open question!)

For LLMs: Noema could potentially serve as the foundation for a language tailored to LLMs. These algorithms act as a description of thought processes, which LLMs could generate, execute, and then critique to iteratively refine their outputs (i.e. thinking path)

I’m looking forward to feedback, ideas, or discussions—any input is welcome! :)