structured-output

Generate JSON, Validate Schema, Transform

The LLM returns prose when you need JSON. Or JSON with a missing field. Or "employees": "five hundred" instead of a number. This pipeline generates structured JSON output, validates it against a schema with required fields and type checks, and transforms the validated data.

Workflow

entity, fields
     │
     ▼
┌──────────────────────┐
│ so_generate_json     │  LLM generates structured JSON
└───────────┬──────────┘
            ▼
┌──────────────────────┐
│ so_validate_schema   │  Check required fields + types
└───────────┬──────────┘
            ▼
┌──────────────────────┐
│ so_transform         │  Transform validated data
└──────────────────────┘

Workers

GenerateJsonWorker (so_generate_json) -- Produces JSON output with a schema definition: required: ["name", "industry", "founded", "employees"] with types map specifying expected types for each field.

ValidateSchemaWorker (so_validate_schema) -- Checks data.containsKey(field) for each required field. Validates type constraints. Returns validation results with pass/fail and error details.

TransformWorker (so_transform) -- Transforms the validated JSON data into the desired output format.

Tests

11 tests cover JSON generation, schema validation with missing/wrong-type fields, and transformation.

Further Reading

• RUNNING.md -- how to build and run this example