Tasks are AI-computed actions that can be launched individually.

There are three types of tasks:

• Extract Meta
• Q&A
• Biblio Search

Extract meta and Q&A are also available from the Knowledge table ('Extract' field). 

Read about extracting knowledge from resources to know more.

TABLE OF CONTENTS

• Extract Meta
  • Configure your Extract Meta
  • Analyse extracted meta
• Q&A
  • Configure your Q&A
  • Analyse answers
• Biblio Search
• Manage tasks
  • Navigate the tasks tab
  • Tasks status
  • Use tasks presets

Extract Meta

Configure your Extract Meta

The Extract Meta feature allows you to retrieve key informations from your documents.

You can request as much information as you need and for each of them you must define:

• the meta needed: for example "authors"
• the format expected: text, whole number (integer), floating number, binary (True/False)
• a guideline (optional): you can for example limit the answer to a specific number of characters or impart the output you are expecting
• multi-valued: you can also specify if the answer can be multiple. For example, if you are requesting for authors names, there can be several, so you need to allow multiple values answer.

Once you validate your configuration, you need to select the file(s) from which you want to extract meta. 

Please note that those meta will only be available in the tasks view for two weeks; if you want save your meta for future operations, you should use the Knowledge table instead.

Analyse extracted meta

Once the task has run, it will appear on the left with a 'success' status.

You can click on the task to access the results.

Each result is sourced and explained. 

You can click:

• on the reference next to the answer to see the in-text justification
• on Explain to get more details about the model reasoning
• on Edit to modify the answer
• on Copy to copy the answer and its reference; you can then paste it to the Editor for example

Q&A

Configure your Q&A

The Q&A feature allows you to question documents and check if the answer is compliant with your expectations.

You can ask as many question as you need and for each of them you must define:

• the question: for example "Is LT3 used as a treatment for hypothyroid?"
• a guideline (optional): for example, you can limit the answer to a specific number of characters or specify the format of the expected output
• compliance check: define the expected output to evaluate whether the answer meets your requirements

Analyse answers

Once the task has run, it will appear on the left with a 'success' status.

You can click on the task to access the results.

Each result is sourced and explained. 

You can click:

• on the reference next to the answer to see the in-text justification
• on Explain to get more details about the model reasoning both for the answer and the compliance result
• on Copy to copy the answer and its reference; you can then paste it to the Editor for example

Biblio Search

The Biblio Search feature allows you to fetch scientific documents online to complete your collections.

To use this feature, you need to:

1. Specify the keyword(s): enter the terms you want to search for. We recommend to use MeSH terms when possible to add accuracy. For example: (disease, hashimoto[MeSH Terms]) AND (hypothyroidism[MeSH Terms])
2. Select a database: choose at least one of the available databases. Current options are Pubmed and arXiv.
3. Set the file limit: indicate the number of documents you want to retrieve
4. Choose a collection: select the collection where the documents will be stored

Documents fetched will be available in your collection and in the details of the task.

Manage tasks

Navigate the tasks tab

The Tasks tab is organized in three spaces:

1. Left panel, here you can:
   • launch a new task
   • search for an existing task (it will match on the preset name)
   • filter by task status (success, failed, running, pending) or type (Q&A, Extract Meta, Biblio Search)
   • access archived tasks
   • see the history of past tasks (up to two weeks)
2. Center panel. This area displays the details of the task selected in the left panel. You can access:
   • the preset name if a preset was used, otherwise the type of task
   • the name of the user who launched the task
   • the timestamp of the task
   • the preset or task details by clicking on the menu button
   • the possibility to archive the task, by clicking on the menu button
   • for an Extract Meta or Q&A: document(s) used and meta or answers with explanation
   • for a Biblio Search: keywords used and collection in which the documents have been stored and list of documents with their database
3. Right panel. This section provides file previews, with highlighted text showing the parts the model used to generate answers for Q&A and Extract Meta tasks.

Tasks status

A task may take some time to be computed, especially if there are many files or many questions to answer.

You can leave the page and work on something else and come back later to find the task ready on the left.

A task can have the following statuses:

• pending: the task has not been treated yet, it is waiting in queue
• running: task is being computed
• success: task is completed and ready for analysis
• failed: task was unsuccessful (if you encounter this use case, please reach out to support)

Use tasks presets

If you often need to ask the same questions or extract the same data, you can use presets.

They allow to load saved configurations and quickly launch a new task.

You can edit your preset with or without saving the changes to adapt your configuration to the current context.

To see how to configure them, take a look at our article Manage presets.