Built-in Functions Reference

The AI Task system provides a variety of built-in functions for common operations in AI workflows. This document serves as a reference for these functions, explaining their parameters and usage.

File Operations

convert_to_docx

Converts a TXT file to DOCX format with specific formatting for transcripts.

Parameters: - source: Path to input TXT file (required) - result: Path to output DOCX file (required)

Example:

- name: convert_to_docx
  type: function
  function: convert_to_docx
  params:
    source: "profile/profile_{{id}}/transcription/INQUA2_{{id}}_transcription_01.txt"
    result: "profile/profile_{{id}}/transcription/INQUA2_{{id}}_transcription_01.docx"

aisource

Loads content from a file into the pipeline.

Parameters: - file: Path to the file to load (required) - dir: Base directory (defaults to current working directory) - content: Optional content to use instead of reading from the file

Example:

- name: load_audio
  type: function
  function: aisource
  params:
    file: "profile/profile_{{id}}/audio/document_{{id}}.m4a"

airesult

Saves pipeline content to a file.

Parameters: - file: Path where the content should be saved (required) - dir: Base directory (defaults to current working directory)

Example:

- name: save_transcript
  type: function
  function: airesult
  params:
    file: "profile/profile_{{id}}/transcription/INQUA2_{{id}}_transcription_01.txt"

Content Processing

content

Extracts or processes text content from the pipeline.

Parameters: - format: Format type (defaults to “text”) - extract: Optional regex pattern to extract specific content

Example:

- name: extract_json
  type: function
  function: content
  params:
    extract: "```json\\n(.*?)\\n```"

count_tags

Counts occurrences of a specific tag in the content.

Parameters: - tag: Tag name to count (required)

Example:

- name: count_speakers
  type: function
  function: count_tags
  params:
    tag: "speaker"

Media Processing

youtube_download

Downloads a YouTube video and adds it to the production.

Parameters: - url: YouTube URL (required) - production_id: Production ID to associate with the video (required) - production_dir: Base production directory (optional) - resolution: Desired video resolution (optional, default: “720p”)

Example:

- name: download_video
  type: function
  function: youtube_download
  params:
    url: "https://www.youtube.com/watch?v=example"
    production_id: "nepi_001"

extract_slides

Extracts slide images from a video based on timestamps in a slide detection file.

Parameters: - slide_file: Path to file containing slide timestamps (required) - video_file: Path to video file (required if not in pipe_data) - output_dir: Directory to save slide images (required) - format: Image format (optional, default: “jpg”)

Example:

- name: extract_slide_images
  type: function
  function: extract_slides
  params:
    slide_file: "profile/profile_{{id}}/slide/INQUA2_{{id}}_slide_01.txt"
    video_file: "profile/profile_{{id}}/video/document_{{id}}.mp4"
    output_dir: "profile/profile_{{id}}/slide/images"
    format: "jpg"

Custom Function Registration

You can register custom functions to extend the system’s capabilities:

from ai_task.funcs import register

def my_custom_function(pipe_data, params):
    # Process data
    result = some_processing(pipe_data.get("pipein_text", ""), params)
    
    # Return updated pipe_data
    pipe_data["pipeout_text"] = result
    return pipe_data

# Register the function
register("my_custom_function", my_custom_function)

Once registered, the custom function can be used in partitur files:

- name: process_data
  type: function
  function: my_custom_function
  params:
    param1: "value1"
    param2: "value2"

Function Placement

Custom functions can be placed in the following locations:

1. Project-level: function/ directory in your project root
2. Profile-specific: profile/profile_{{id}}/function/ for profile-specific functions

Example project structure:

project/
├── function/                    # Project-level functions
│   ├── __init__.py
│   ├── custom_docx.py
│   └── custom_audio.py
├── profile/
│   ├── profile_001/
│   │   ├── function/           # Profile-specific functions
│   │   │   ├── __init__.py
│   │   │   └── custom_format.py

The system searches for functions in the following order: 1. Built-in functions in the AI Task package 2. Project-level functions in function/ 3. Profile-specific functions in profile/profile_{{id}}/function/

This allows profile-specific functions to override project-level functions, and project-level functions to override built-in functions, providing a flexible customization system.