DSPy Module Builder

DSPy (Declarative Self-improving Python) is a framework from Stanford that replaces hand-crafted prompts with learnable modules. Instead of writing brittle prompt strings, you define typed Signatures that describe what you want, then let DSPy compile and optimize those prompts automatically using your examples and a chosen metric.

A Signature is a declarative spec of a module's I/O contract. It subclasses dspy.Signature and defines input/output fields as class attributes using dspy.InputField() and dspy.OutputField(). The docstring acts as the task instruction. DSPy compiles this into an optimal prompt under the hood.

Use dspy.Predict for fast, direct tasks where reasoning is not needed (e.g. classification, extraction). Use dspy.ChainOfThought when the task benefits from step-by-step reasoning, such as math problems or multi-hop QA. ChainOfThought adds an intermediate 'rationale' field that guides the model.

dspy.ReAct implements the Reasoning + Acting loop. It interleaves LLM reasoning steps with external tool calls (search, code execution, APIs). Use it when your pipeline needs to query databases, call APIs, or interact with external services to answer questions.

After defining your module, collect a small labeled dataset and pick an optimizer like dspy.BootstrapFewShot or dspy.MIPROv2. Call optimizer.compile(module, trainset=your_data) and DSPy will automatically search for the best prompts and few-shot examples, saving you manual prompt engineering effort.