pydantic.md

Data Validation & Schemas (Pydantic)

I use Pydantic v2 across my project to enforce strict type checking, automatic runtime validation, and structured data handling for LLM inputs and outputs.

Why Pydantic v2?

• Performance & Core Engine: Built on Rust for extremely fast parsing and validation.
• Modern API: Utilizes standard V2 idioms like model_validate() and model_dump() rather than deprecated V1 methods (parse_obj(), .dict()).
• Python 3.9+ Type Hints: Integrates seamlessly with standard generic hints (list[str], dict[str, Any]) without requiring legacy typing.List or typing.Dict wrappers.

Implementation & Request Modeling

I define structured BaseModel classes to represent standard API request payloads, system tool definitions, and parsed LLM responses:

from typing import Any
from pydantic import BaseModel


class Message(BaseModel):
    role: str
    content: str


class FunctionParameter(BaseModel):
    type: str
    properties: dict[str, Any]
    required: list[str]


class Function(BaseModel):
    name: str
    description: str
    parameters: FunctionParameter


class Tool(BaseModel):
    type: str
    function: Function


class RequestData(BaseModel):
    model: str
    messages: list[Message]
    tools: list[Tool]
    stream: bool = False
    think: bool = False

How I Use Pydantic in My Pipeline

• Guaranteed Schema Adherence: Every JSON response generated by the LLM is validated against models like SelectedFunction.
• Runtime Error Boundary: If the model outputs invalid parameters or missing required fields, Pydantic triggers a ValidationError. My parsing logic captures this failure cleanly, returning a default null object instead of allowing unhandled runtime exceptions to crash the application.
• Serialization: Simplifies conversions between Python dictionaries, raw JSON strings, and typed objects via .model_dump() and .model_dump_json().

References

• Pydantic v2 Documentation
• PyPI: Pydantic Package