Summary

The “Document Tasks” application in DeepThought makes it easy to process documents or spreadsheet data in bulk. This can be useful for activities such as grading assignments, extracting information from several documents, classifying content, or generating structured outputs from spreadsheet data. 

 

Accessing the Document Tasks Application 

1. In DeepThought (deepthought.usnh.edu), open the Applications menu. 

2. Select Document Tasks. 

3. To create a new task, click Create document task in the left sidebar. 

 

Creating a Document Task 

When creating a task, complete the following configuration steps: 

 

1. Name the task 

Enter a descriptive name that clearly identifies the purpose of the task. 

Examples: 

• Student Quiz Grading

• Resume Analysis

 

2. Select the input type 

Choose the type of content the task will process.

 

Option 1: Collection of documents 

Use this option when you want the task to analyze one or more text documents independently. 

You can upload: 

• Individual documents, such as a word or pdf files. 

• A ZIP file containing multiple documents. 

 

Option 2: Rows in a spreadsheet 

This option analyzes spreadsheet data one row at a time. 

• Important notes: 

• Each row is processed independently. 

• The task cannot compare rows to one another. 

• The task cannot sort, modify, insert, or create spreadsheet rows in the uploaded file. 

• The task only evaluates the information contained within the current row being processed. 

 

3. Select the output type 

Choose the format the task results will be returned in. There are two options, “Spreadsheet” or “Custom Text-based Data.” 

Spreadsheet (recommended) 

• This is the default and preferred option for most use cases. 

• Benefits include: 

• Easy review of results 

• One output row per processed item 

• Simple export and download capabilities 

• Compatibility with common spreadsheet applications

Custom text-based data 

• This specialized option is primarily intended for advanced use cases involving custom code or structured text outputs. 

 

4. Select a chat 

Choose the custom chat that the task will use when processing content. 

The selected chat determines: 

• The language model used. 

• Temperature and response settings. 

• Available data sources. 

• Uploaded reference documents. 

• Web search permissions. 

• Other chat-specific configuration settings. 

For example, a user might create a chat named “My Grading Task Chat.” Documents containing answer keys or grading rubrics could be added to that chat's settings, and web search could be disabled. Any document task using that chat would then have access to those resources and settings. 

For more information about configuring chats, see: 

DeepThought Chatbot Settings and Administration 

 

5. Add per-item instructions 

Per-item instructions tell the task application what to do with each document or spreadsheet row. 

These instructions are applied to: 

• Every uploaded document when using the “Collection of Documents” input type 

• Every spreadsheet row when using the “Rows in a Spreadsheet” input type 

Example: 

A professor grading student quiz submissions might enter these instructions: 

“Grade each uploaded document. Add the document name in the first column, the score in the second column, and grading notes in the third column.” 

If grading rubrics or answer keys have been added to the selected chat, the task can use those resources when performing the evaluation. The resulting output spreadsheet would contain one row for each uploaded document. 

 

Viewing and Downloading Results 

After processing is complete: 

1. Review the results in the task output preview. 

2. Click the download icon to export the results. 

3. Open the downloaded file in the spreadsheet application of your choice. 

 

Important Considerations 

File upload requirements 

• Choose the input type that matches the file type of the content you are uploading. 

• Only one file can be uploaded at a time. 

• To upload multiple documents at once, place them into a ZIP file before uploading. 

• All files within a ZIP archive must be the same type, either documents or spreadsheets. 

 

Spreadsheet processing limitations 

When using spreadsheet input: 

• DeepThought processes each row independently. 

• The application cannot analyze the spreadsheet as a whole. 

• The application cannot compare rows, calculate relationships between rows, sort data, or add rows. 

 

When writing instructions for spreadsheet processing, avoid directions such as: 

• "Create a spreadsheet" 

• "Insert rows" 

• "Compare all rows" 

• "Sort by score" 

Instead, focus instructions on analyzing or transforming the data contained in each individual row. 

 

Output selection 

The selected output type determines how results are returned. 

For most workflows, Spreadsheet output is recommended because it provides structured results that are easy to review, download and further analyze. 

 

Performance considerations 

Large spreadsheets may require significant processing time because rows are analyzed individually. 

When working with large datasets: 

• Consider breaking large spreadsheets into smaller batches. 

• Process each batch separately to improve manageability and reduce wait times. 

 

Spreadsheet best practices 

For best results: 

• Use clear, descriptive column headers. 

• Ensure per-item instructions reference the correct column names. 

• Verify spreadsheet data is well organized before uploading. 

 

Usage of premium models and web search 

If the selected chat uses a premium model such as Claude and has web search enabled, task processing can consume credits rapidly. 

Consider: 

• The size of the task. 

• The number of files being processed. 

• Other DeepThought activity occurring within the same 24-hour credit period. 

 

Modifying task behavior 

Changes to language models, data sources, web search permissions and related settings must be made by updating the selected chat configuration. 

Changes to the actual processing steps or workflow should be made in the Per-Item Instructions field within the Document Task settings.

 

Further Readings

USNH AI Chatbots: Overview

DeepThought: Chatbot Settings and Administration

USNH AI Tools: Understanding Data Tiers

 

Want to Learn More? 

Fill out the RCC Support Request form.

 

Need additional help?

For guidance on this and other related topics, register through Workday to join any of our upcoming Drupal/Sites@USNH/DeepThought/Website Tools - Virtual Walk-in Sessions.

If you have any additional questions, please fill out the Website & Mobile Development Support Form.