Parser¶

The parser defines the output shape: the fields, their types, and nesting. Internally this is a Pydantic model.

You can create it in two ways:

using the visual builder (recommended): build-parser
writing the Pydantic model by hand

1. Visual builder¶

Run:

build-parser

This opens the UI where you can:

add fields (string, int, float, list, nested model)
rename the model
preview the generated Python
export it

When you export, save the file to:

tasks/parsers/<name>.py

2. Structure of the generated file¶

A generated parser file usually looks like:

from pydantic import BaseModel
from typing import Optional, List

class OutputParser(BaseModel):
    patient_id: str
    findings: Optional[str] = None
    measurements: Optional[List[float]] = None

You can edit this file manually if you want to add validators or docstrings.

Tip

You can also write your own Pydantic models from scratch. Just make sure they inherit from BaseModel and the top level class is named OutputParser. You can ask your favorite LLM to help you with setting it up for your specific use case!

3. Using the parser in a task¶

In your task JSON:

{
  "Parser_Format": "report.py"
}

The extractor will load the Python model from tasks/parsers/report.py and tell the LLM to return JSON that matches it.

4. Good practices¶

keep field names lowercase and descriptive
prefer Optional[...] for fields that might not be present in the text
start with a small model and expand it once the LLM returns consistent data