dspy-ruby

Type-safe LLM applications in Ruby use Sorbet-based signatures to define structured inputs and outputs, ensuring predictable JSON schema interactions with language models. Modular components and provider adapters combine to create robust production workflows.

Structuring modular LLM workflows involves composing predictors, tools, and modules using fiber-local LM contexts. This allows you to chain components together while maintaining observability and optimization throughout the lifecycle of the request.

To get predictable JSON outputs from an LLM in Ruby, define type-safe signatures that map to JSON schemas. These signatures enforce structured responses, making your LLM interactions robust and reliable for downstream processing.

Yes, Sorbet is required because DSPy.rb relies on Sorbet-based signatures to enforce type safety. You need to set up a Ruby environment with the dspy gem and define signatures to wire models, tools, and evaluators.

Yes, you can chain multiple LLM predictors and tools together by composing modules with features like with_lm and lifecycle callbacks. This supports building end-to-end LLM apps with complex chaining and evaluation workflows.

Using type-safe signatures requires strict adherence to defined JSON schemas, which may limit flexibility for highly open-ended or unstructured generation. You must carefully wire models, tools, and evaluators as described in the documentation to avoid workflow mismatches.