Build a JSON Schema for Structured AI Outputs

Structured output APIs from OpenAI (json_schema mode) and Anthropic (tool use) require a JSON Schema definition that constrains the model's response format. When you provide a schema, the API guarantees that the response matches it — eliminating the need for JSON repair and making downstream parsing reliable. This example shows how to build a schema for a common use case: extracting structured information from unstructured text. The schema builder generates schemas compatible with both OpenAI's strict mode and Anthropic's tool definition format. Key differences: OpenAI strict mode requires additionalProperties: false on all nested objects and does not support all JSON Schema keywords. Anthropic tool definitions use the same field names but are wrapped in a tool definition object. The builder outputs both formats from a single schema definition. When designing schemas for LLM extraction, prefer flat structures over deep nesting — models fill in required fields more reliably when the schema is simple. Use enum constraints to force the model to choose from a predefined list rather than generating arbitrary strings. Mark fields as required rather than optional wherever possible, since models often skip optional fields even when the source text contains the relevant information.

Extract the following fields from a job posting:
- title (string, required)
- company (string, required)
- location (string, required)
- salary_range (object with min and max integers, optional)
- employment_type (enum: full-time, part-time, contract, required)
- remote (boolean, required)
- required_skills (array of strings, required)
- years_experience (integer, optional)