Skip to main content

​
Overview

Multi-modal Knowledge Extraction is a feature that extracts information from unstructured data sources by processing multiple data modalities simultaneously, including text, images, structured data, and visual content. This feature enables comprehensive knowledge extraction from complex, heterogeneous data sources.

​
Category

Unstructured Data - This feature is designed to work with unstructured data sources such as documents, PDFs, images, and other non-tabular formats.

​
Input

The Multi-modal Knowledge Extraction feature accepts data from the following input sources:
  1. File Upload - Upload unstructured files directly from your local system. Supported formats include PDFs, images, text files, and documents with embedded content.
    Learn more: Data Sources - Upload and manage data sources
  2. Live Data Connectors - Connect directly to live data sources without duplicating data:
    • Blob Storage: Amazon S3, Azure Blob Storage, Google Cloud Storage (GCS) buckets
    • Databases: Snowflake, Databricks, and other database systems
    Learn more: ▶️ Live Data Connectors Tutorial - Connect directly to live data sources
The input data source must be marked as “Unstructured Type” in your data room.

​
Output

Corvic Table - The Multi-modal Knowledge Extraction feature produces a new Corvic Table containing:
  • Extracted Text Table: Structured text content extracted from documents
  • Image Table: Image-related information and metadata
  • Knowledge Entities: Identified entities such as people, organizations, locations, dates, products, and concepts
  • Relationships: Connections and relationships between extracted entities
  • Structured Information: Additional structured data derived from the input data
The output Corvic Table provides a comprehensive structured representation of all knowledge extracted from your multi-modal input data, ready for use in downstream analysis, embeddings, or agent workflows.

​
Parameters

​
Usage Example

To use Multi-modal Knowledge Extraction in a Workflow:
  1. Add your unstructured data source to the Workflow canvas
  2. Click the ”+” button next to the data source
  3. Select “Multi-modal Knowledge Extraction” from the actions menu
  4. Select the input data source (if not already selected)
  5. Optionally provide a name for the output Corvic Table
  6. Run the Workflow to execute the extraction
  7. Review the generated Corvic Table containing extracted text table, image table, knowledge entities, relationships, and structured information

Workflows

Learn how to build workflows using Workflows.

Corvic Tables

Understand how Corvic Tables work with extracted knowledge.

Spaces

Learn about embedding spaces generated from extracted knowledge.

Data Sources

Learn how to upload and manage unstructured data sources.