dspy

You can stop manually tweaking prompts by using declarative programming with DSPy. Define tasks as input-to-output signatures and apply optimizers like BootstrapFewShot to automatically improve prompt quality using labeled examples and evaluation metrics.

Automatic prompt optimization works by using teleprompters like MIPRO over your modular pipelines. You provide a metric and labeled data, and the optimizer selects effective few-shot examples and instructions to improve RAG, classification, and multi-stage agent performance.

Yes, you can use DSPy with OpenAI and Anthropic models. You configure a DSPy language model provider for your chosen service, allowing you to build and optimize your declarative language-model programs across different supported LM providers.

The best way to build multi-stage systems is using declarative programming with structured signatures. You compose modular predictors like ChainOfThought and ReAct alongside retrieval modules to create maintainable pipelines for complex reasoning and extraction tasks.

Yes, you need labeled data to optimize language model programs. Optimizers such as BootstrapFewShot require a defined evaluation metric and a set of labeled examples to automatically select prompts and improve output quality.

Declarative signatures replace manual prompt engineering with clear input-to-output contracts, making your pipelines more maintainable. This structured approach allows optimizers to systematically improve performance rather than relying on inconsistent manual tweaks.