dspy

Automate prompt optimization by applying teleprompters like BootstrapFewShot, MIPRO, and BootstrapFinetune to iteratively improve prompt instructions and few-shot examples using labeled data, replacing ad-hoc manual tuning with data-driven accuracy boosts.

Build modular RAG pipelines by composing declarative modules like ChainOfThought and ReAct to express multi-stage reasoning tasks as structured input to output transforms, enabling reliable retrieval, reranking, and verified answer generation.

Declarative LM programming supports multiple model providers including OpenAI, Anthropic, and local models, allowing you to define signature-based interfaces and compose complex workflows across different platforms without changing your core pipeline logic.

Build tool-using agents by composing modular components like ReAct and retrievers within a declarative programming framework, enabling your agent workflows to retrieve relevant passages, rerank candidates, and execute multi-hop question answering reliably.

Inconsistent classifier outputs often stem from brittle ad-hoc prompt engineering; defining structured Signatures and applying automated teleprompters like BootstrapFewShot standardizes the input to output transforms and optimizes few-shot examples for reliable accuracy.

You can fine-tune and optimize prompts using your own labeled question-answer examples by applying the BootstrapFinetune teleprompter, which leverages your data to automatically improve and adapt few-shot demonstrations for better task performance.