DSPy
dspy.ai
A Stanford-developed framework for algorithmically optimizing LLM prompts and pipelines, replacing hand-written prompts with compiled programs that automatically improve based on defined metrics.
Why it is useful
Treats prompt engineering as an optimization problem rather than a craft. Instead of manually tweaking prompt wording, you define your pipeline and a metric for success, and DSPy finds the prompt instructions and few-shot examples that maximize it. Most useful when you have evaluation data and want reproducible, measurable prompt quality rather than prompts that feel good intuitively.