schema-guided-reasoning

What problem does it solve?

Schema-Guided Reasoning (SGR) provides a pattern to guide LLMs through a predefined structure, enabling reproducible, auditable, and testable reasoning by constraining outputs to a typed schema.

Core Features & Use Cases

• Structured outputs: define schemas with Pydantic models and use response_format/parse to enforce a fixed result structure.
• Deterministic reasoning order: the sequence of fields drives the model's step-by-step thinking.
• Auditable & testable: intermediate fields and outputs can be validated and reviewed.
• Use cases: evaluation, risk assessment, compliance verification, and systematic data extraction across domains.

Quick Start

Define a Pydantic schema and show how to parse a model response with response_format.