Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

...

In custom workspaces, workflows can be visually designed from the available workflow resources on the right. Workflow resources can be created using right-click or via the left panel for on the workflow editor or in the case of collections and analysis using the left panel.

DnDSteps.gif

image-20240905-093050.png

Document Analysis

In the document collections (1) [todo: Separate screenshot?], you see all the documents you uploaded. When opening a document there, you’re presented with the analysis view, which allows you to configure the analysis and its steps itself (2), as well as interact with the document (3) by editing and annotating the content directly on the page.

...

Collections

Collections are folders storing your documents for the workflow setup and exploration phase.

Analyses

An essential part of almost every workflow are analyses. In the analysis you can configure which steps should be run on your documents with the following main purposes:

  1. Pre-process

  2. Structure

  3. Extraction

  4. Post-process

For most of the steps you can visually interact with your documents. This enables you to specify your workflow according to your individual use-case and train a tailored AI-model.

From the collections you can open documents in the analysis document viewer featuring from the left to right:

  • The steps control panel

  • A preview of the document with options to interact with the selected step

  • Various previews for the extracted data

...

Model evaluation

For each analysis, the steps that are trainable generate models, which you can create and manage here. You can evaluate the model performance, decide which model to use as the default active one, add notes and delete them if no longer needed.

Todo: make gif clicking on evaluation showing detailed F1 scores.

Evaluation.gifImage Removed

image-20240905-102858.pngImage Removedmodels. For each trainable step in the analysis an separate tailored model is trained on the selected training set using the document state as provided by the previous steps. The performance of each model can be evaluated for each step individually.

...