dspy

Declarative programming enables self-improving AI pipelines by composing LM calls and optimizers into modular components. This eliminates manual prompt tuning by using built-in teleprompters to automatically optimize multi-stage ML workflows.

Declarative module composition is a method to build end-to-end AI systems using components like Predict, ChainOfThought, ReAct, and ProgramOfThought. It allows you to structure complex reasoning steps and LM calls without writing imperative prompt instructions.

You set up a multi-stage RAG pipeline by declaratively composing retrieval, thinking, and answering steps. Built-in optimizers then automatically refine the language model calls within these stages using your provided data.

Yes, you can build multi-hop QA agents and classifiers using declarative module composition. The framework supports agents and multi-stage pipelines that improve over time through automatic optimization workflows.

No, declarative programming removes the need for manual prompt engineering by using optimizers. These teleprompters automatically tune the language model calls within your classifiers and RAG systems based on data.

Declarative LM pipelines require defining modular components and providing data for the automatic optimization process to function effectively. Teams must structure their multi-stage ML workflows declaratively rather than relying on isolated prompt adjustments.