Dspy

If you're tired of manually tweaking prompts and want a more systematic approach to building AI pipelines, this is worth checking out. DSPy comes from Stanford NLP with a solid 22k+ GitHub stars and treats language model programming like actual programming, not prompt witchcraft. You declare what you want, then let optimizers tune the prompts using data instead of guesswork. It's built for complex stuff like RAG systems, agents, and multi-step workflows where you need components that actually compose well. The modular approach means you can iterate without rewriting everything, which is the whole point. Think of it as bringing software engineering discipline to LM applications.