chicagopeabodydev-sudo/pydantic-skill

What are Pydantic models and when to use

Pydantic models group related fields into a single component. In RAG applications, they define structured output—the expected format the LLM returns (e.g., answer plus citations, confidence scores). They can also define structured input for tool calls or extraction pipelines.

Pydantic models: name and collection of fields

Each field needs at minimum:

1. Name of the field
2. Type of data (e.g., int, str, List[str])

Nested models

Pydantic models can contain other models as fields:

from typing import List
from pydantic import BaseModel, Field

class Song(BaseModel):
    """Data model for a song."""
    title: str
    length_seconds: int

class Album(BaseModel):
    """Data model for an album."""
    name: str
    artist: str
    songs: List[Song]

Defining models for structured output (RAG / LlamaIndex)

1. Name the model (PascalCase)
2. Docstring describing what the model represents (helps the LLM)
3. Fields in the form field_name: type
4. Optional: Field(..., description="...") to guide the LLM on what to populate

summary: str = Field(
    ..., description="A concise summary of this text chunk."
)

Using with LlamaIndex

• Query responses: Use llm.as_structured_llm(output_cls=YourModel) and pass the structured LLM to index.as_query_engine(llm=sllm) so RAG queries return Pydantic objects.
• Extraction: Use PydanticProgramExtractor or OpenAIPydanticProgram for extracting structured data from documents.

Additional Resources

• For usage examples, see examples.md
• Pydantic BaseModel documentation
• LlamaIndex structured outputs