DSPy

DSPy takes a different approach to building with language models. Instead of tuning prompt text by hand, you write Python code that declares what each step should do, and DSPy figures out the prompts, examples and even model weights using an optimizer. You define signatures, combine them into modules and give the system some examples and a metric. It then searches for a version of the program that scores best, which helps when you switch models or change the task. DSPy is free and open source, born at Stanford. It suits developers who want a more systematic, testable way to build LLM pipelines.