DSPy

DSPy is an open-source framework from Stanford NLP for building LLM-powered systems through composable Python code rather than manual prompt engineering. It provides three core abstractions: signatures (input/output specs), modules (reusable components), and optimizers that automatically synthesize effective prompts and few-shot examples. DSPy can also fine-tune model weights. It supports building classifiers, RAG pipelines, agent loops, and other LLM applications while making them more reliable and maintainable than hand-crafted prompts.