Skip to content

Manual Configuration

This page is for users who want to create task files and parsers by hand without the Studio.

Directory layout

A common layout is:

.
├── data/
│   └── reports.csv
├── tasks/
│   ├── Task001_reports.json
│   └── parsers/
│       └── report.py
└── output/
  • data/ — your source CSV/JSON
  • tasks/ — your task JSONs
  • tasks/parsers/ — Python files with Pydantic models
  • output/ — where extraction results go

Task JSON fields

Minimum viable task:

{
  "Description": "Extract structured info from radiology reports",
  "Data_Path": "data/reports.csv",
  "Input_Field": "text",
  "Parser_Format": "report.py"
}

The only additional field you may want is Example_Path if you intend to use example-based prompting.

Naming

Use the TaskXXX_name.json style so the CLI can pick task IDs reliably:

  • Task001_reports.json → --task_id 1
  • Task010_products.json → --task_id 10

Stick to integers in the CLI to avoid confusion.