dspy

Declarative language model programming builds modular AI pipelines using signatures and composable components to optimize prompts automatically. It replaces manual prompt engineering with data-driven methods, improving model behavior systematically using evaluation metrics and training examples.

You optimize LLM prompts automatically by configuring optimizers with metrics and training examples. This evaluation-driven approach systematically improves model behavior for RAG systems, classifiers, and multi-stage pipelines without manual prompt tuning.

Yes, you can build multi-stage AI agents with tool-using capabilities. Declarative programming supports composable modules, allowing you to construct modular workflows and pipelines that integrate tool usage and multi-stage processing.

The best way to move beyond manual prompt engineering for RAG applications is using data-driven optimization. Declarative programming applies optimizers and metrics to systematically improve RAG pipelines from evaluation data rather than manual adjustments.

Yes, language model provider integration is required to optimize AI workflows. Configuring provider integration enables modular programming, optimizer configuration, and metric-driven improvement workflows to function across your LLM application development.

You should avoid data-driven prompt optimization when lacking evaluation data or clear metrics. Without training examples to feed optimizers, the modular programming approach cannot systematically improve prompts or validate model behavior improvements.