Search Results

Generative programming is a paradigm that takes decades of theoretical research and practical experience in algorithms and software engineering and applies it to the way we interact with LLMs. Instead of developing prompts by trial and error, which usually results in long and complex prompts, generative programs combine known and well-tested control flows and design patterns, such as divide and conquer, with short, consumable prompts. I will describe Mellea, an open-source library for writing generative programs, replacing brittle prompts with structured, maintainable, robust, and efficient AI workflows. I will then discuss some challenges and proposed solutions for efficiently handling multiple models and LLMs’ internal memory management within a generative program.