Skip to main content

Prediction Overview

Improve AI model accuracy by creating training data through manual labeling and leveraging machine learning predictions for automated document processing.

Core Concepts​

Training Data Creation​

Build high-quality datasets by labeling task files to train AI models for your specific document types:

  • Manual Labeling - Apply labels directly to uploaded task content
  • Training Marks - Flag correctly labeled tasks as training data
  • Label Types - Different labeling approaches for each processing step

AI Model Integration​

Import and configure AI models for automated prediction across your processing pipeline:

  • Model Import - Load pre-trained models via Pipeline settings
  • Label Mapping - Connect project labels to model labels
  • Step-Specific Models - Different model types for each processing stage

Automated Prediction​

Generate predictions using configured AI models and review results before finalizing:

  • Auto Predict - Automatically predict new uploads
  • Manual Prediction - Trigger predictions on-demand
  • Label Removal - Clear predictions and start fresh

Processing Steps​

Different steps in your processing pipeline use distinct labeling and prediction approaches:

Classification​

Categorize documents into types and route to appropriate next steps:

  • Classification Labels - Define document types and routing rules
  • Hotkey Assignment - Quick labeling with keyboard shortcuts
  • Threshold Configuration - Set confidence levels for automatic processing

Segmentation​

Divide document pages into logical segments for targeted processing:

  • Segment Creation - Define content boundaries manually or automatically
  • Right-Click Editing - Remove or modify segments with context menus
  • Coordinate-Based - Visual segment definition on document pages

Extraction​

Annotate and extract specific data fields from documents:

  • Entity Labels - Define data fields to extract
  • Annotation Placement - Mark specific text or content areas
  • Field Descriptions - Help AI models understand extraction targets

Grouping​

Split document pages into separate groups or combine related content:

  • Context Grouping - Organize related pages together
  • Group Labels - Define grouping criteria and relationships
  • Model-Assisted - Use AI to identify natural groupings

Training Workflow​

1. Label Creation​

Create labels specific to each processing step in your pipeline. Each step type has different labeling requirements and approaches based on the content being processed.

2. Manual Labeling​

Apply labels to uploaded task files:

  • Place labels manually on content
  • Right-click to remove incorrect labels
  • Use keyboard shortcuts for quick labeling

3. Training Marks​

Mark correctly labeled tasks for training data:

  • Use "Mark for training" button for approved tasks
  • Build datasets across different document types
  • Quality control through manual review

4. Model Configuration​

Import and configure AI models in Pipeline settings:

  • Map project labels to model labels
  • Set prediction thresholds
  • Configure automatic processing rules

5. Prediction & Review​

Generate and review AI predictions:

  • Use "Predict" button for manual predictions
  • Enable auto-predict for new uploads
  • Use "Erase" to clear all labels and restart

AI Model Management​

Model Import​

Access model management through Project Settings → Pipeline:

  1. Select the processing step to configure
  2. Choose "Model" in automation settings
  3. Import models from available options
  4. Map project labels to model labels

Label Mapping​

Ensure proper label alignment between your project and imported models:

  • Project labels define your extraction fields
  • Model labels represent AI training categories
  • Mapping connects project needs to AI capabilities
  • Mismatched labels can cause prediction errors

Model Types​

Different AI model types serve specific processing needs:

  • NER Models - Named Entity Recognition for extraction
  • Classification Models - Document type categorization
  • Segmentation Models - Content boundary detection
  • Grouping Models - Context-based page organization

Related Documentation​

Under Development

The AI model system is actively under development. Features and interfaces may change as new capabilities are added.

Key Features​

  • AI model training and deployment
  • Prediction accuracy optimization
  • Custom model configurations
  • Performance monitoring

Getting Started​

Learn about training models and deploying AI predictions for your document processing workflows.