Workflows

Under Review

This article is currently under review. Some content may be incomplete or inaccurate.

Workflows define the processing pipeline for a project. They determine which activities run and in what order.

Workflow Structure

A workflow is a sequence of activities:

{
  "activities": [
    { "type": "import", "enabled": true },
    { "type": "data_transform", "enabled": true },
    { "type": "split", "enabled": true },
    { "type": "classify", "enabled": true },
    { "type": "extract", "enabled": true }
  ]
}

Activity Types

Type	Purpose	Input	Output
import	Ingest uploaded files	Raw files	Registered documents
data_transform	Convert to processable format	Documents	Page images + OCR
split	Detect document boundaries	Pages	Document groupings
classify	Categorize documents	Document text	Classification labels
extract	Pull structured data	Document text + schema	Field values

Configuring Workflows

Enabling/Disabling Activities

Toggle any activity on or off in the project settings. For example, if your documents are already single-page and pre-classified:

{
  "activities": [
    { "type": "import", "enabled": true },
    { "type": "data_transform", "enabled": true },
    { "type": "split", "enabled": false },
    { "type": "classify", "enabled": false },
    { "type": "extract", "enabled": true }
  ]
}

Activity Configuration

Each activity can be configured independently:

Classification Config

Setting	Description	Default
model	Azure OpenAI model to use	gpt-4.1-mini
temperature	Model creativity (0-1)	0.0
confidence_threshold	Minimum confidence score	0.7

Extraction Config

Setting	Description	Default
model	Azure OpenAI model to use	gpt-4.1-mini
temperature	Model creativity (0-1)	0.0
include_coordinates	Return bounding boxes	true
confidence_threshold	Minimum confidence score	0.7

Settings Inheritance

Configuration follows an inheritance chain:

Project Defaults → Workflow Activity Config → Runtime Overrides (per transaction)

This lets you set sensible defaults at the project level while allowing per-transaction customization when needed.

Workflow Execution

The workflow engine:

1. Picks up transactions from the Redis queue
2. Runs each enabled activity in sequence
3. Tracks progress and status per activity
4. Handles retries for transient failures
5. Reports completion or failure

Monitoring

View workflow progress in the transaction detail view. Each activity shows:

• Status: Pending, running, completed, or failed
• Duration: How long the activity took
• Errors: Detailed error messages if something went wrong