Definition
Plain language
A framework for building and automatically tuning multi-step AI pipelines by treating prompts like code you can optimize.
As stated in the literature
A programming framework that compiles and optimizes chains of LLM calls, with modules like MIPROv2 for prompt optimization; a standard baseline in LLM-pipeline optimization work.
Why it matters: It matters because it turns fragile hand-written prompt chains into something you can systematically build, test, and tune like real code.
For example, instead of hand-tweaking the wording of a multi-step prompt, you let DSPy automatically search for phrasings that produce better answers.
Heard on the show
“… And existing optimizers for these systems — TextGrad, DSPy, GPTSwarm, that whole lineage of treating an LLM pipeline as a program you can automatically improve …”Episode 181 — How to Backpropagate Blame Through a Team of Chatbots — And When It Backfires