Skip to main content

AI Find By Question

Overview​

AI Find By Question extracts answers from documents using questions you define in natural language. You select files or folders, define questions (each one becomes a column), and review results in a table you can export as CSV or save to a Workspace. Inputs are indexed documents from your workspace. Outputs are a structured results table, optional tags applied to selected items, and exported CSV files. AI results require review, especially when questions are vague or when documents contain multiple possible answers.

When to Use This Application​

  • You need to extract the same fields from many documents (for example vendor name, contract value, effective date).
  • You want a structured table for review, triage, or downstream analysis instead of reading documents manually.
  • You need to tag documents based on extracted values (for example “High value contract”).
  • You want to reuse a standard set of questions across cases using templates.
  • You need a CSV export for sharing or reporting.

Before You Begin​

  • Ensure your documents are available in Octostar and have indexed content.
  • Confirm you have access to the source Workspace and the destination workspace if you plan to save templates or export results to a workspace.
  • Decide what questions you need answered and what output type each answer should take (for example text, numeric, categorical).

Step-by-Step Walkthrough​

Step 1 — Begin extraction​

When the app opens, you see a welcome screen that summarizes the workflow. Select Begin Extraction to start.

Step 2 — Select documents​

Drag and drop documents from the Octostar sidebar into the drop zone. You can add:

  • Individual files
  • Entire folders

After you add documents, the app shows a list with each document’s name and type. Use the controls to manage the selection:

  • Search for a document using the search bar
  • Remove a document you do not want to include
  • Review skipped files (documents without indexed content show a warning)

When your selection is ready, select Next: Add Questions.

Step 3 — Configure questions​

Define the information you want extracted from each document. Each question becomes one column in the results table.

Add and manage questions​

Use the question controls to build your set:

  • Select Add Question to create a new question tab
  • Enter a question in natural language (for example “What is the company name?”)
  • Press Enter in the prompt field to add another question
  • Rename a column tab using the pencil icon
  • Delete a column using the trash icon on the tab

Choose an output type​

Select an output type for each question. The app tries to auto-detect the best type, but you can change it. A description of the selected type appears below the selector.

Configure categorical questions​

If you choose the categorical type, define at least two categories. You can:

  • Add categories manually (name + optional description), then select Add
  • Select auto-detect to have the AI suggest categories from a sample of your documents
  • Remove categories using the X on a category tag

Allow multiple values​

For free text and numeric questions, you can enable Allow multiple values when a document may contain more than one valid answer. Results appear as semicolon-separated values. Examples:

  • “List all authors”
  • “What are the line item amounts?”

Save and load templates​

Use templates to reuse question sets:

  • Save the current questions as a template to any writable workspace
  • Load a saved template to populate the questions instantly Templates are stored as .aiquestions files in your workspace and can be shared with your team.

Adjust advanced settings​

Select Settings to tune performance:

  • Batch Size — How many documents the AI processes in one batch
  • Smaller batches improve accuracy but take longer. Larger batches run faster but may miss details.
  • Parallelism — How many batches run at the same time
  • Higher values increase speed without changing accuracy.

Test before full extraction​

Select Test Extraction to run a trial on the first 5 documents. Use the test run to verify:

  • Results look correct
  • Output types match the answers you expect
  • Questions are specific enough After testing, refine questions if needed. When you are ready, select Start Full Extraction.

Step 4 — Review selection and run full extraction​

Before the run starts, you see a checklist of documents. Use the controls to manage the run set:

  • Select or deselect individual documents
  • Use Select All and Deselect All for bulk changes
  • Review the summary for document count and question count To start, select Run Full Extraction.

Step 5 — Monitor extraction and review results​

While extraction runs, the app shows:

  • A progress bar
  • Elapsed time and estimated time remaining
  • A live results table that updates as documents complete You can select Pause to stop processing temporarily. When paused, you can review partial results, export what is complete, or select Resume to continue.

Step 6 — Tag selected entities (optional)​

After you review results, you can tag items directly from the table.

  1. Select rows using the checkboxes.
  2. Select Tag Selected Entities.
  3. Choose a workspace and select an existing tag, or create a new tag with a name and color.
  4. Select Apply Tag.

Use this to categorize documents based on extracted results (for example “Contracts > 100k EUR” or “Needs review”).

Understanding the Output​

The primary output is an interactive results table. Each row represents a document. Each question you defined becomes a column.

The table supports:

  • Sorting and filtering by any column
    • Text search for free-form columns
    • Checkbox filters for categorical and simple question columns
  • Resizing columns by dragging borders
  • Renaming columns using the edit icon in a column header
  • Opening a document by selecting its name
  • Selecting rows with checkboxes for bulk actions (export or tagging)

Exports include:

  • Document name
  • All question columns
  • Metadata fields such as entity ID, entity type, creation date, and creator

Saving and Exporting Results​

Select Save Table to export results. You can export in two ways:

  • Download CSV — Saves the CSV file to your computer.
  • Save to Workspace — Uploads the CSV into an Octostar workspace so you can share it with your team.

Before saving to a workspace, select:

  • Destination Workspace

Tips for Best Results​

  • Write specific questions. Include units and constraints (for example “total contract value in EUR”).
  • Run Test Extraction before a full run to validate question wording and output types.
  • Start with a smaller Batch Size when accuracy matters, then increase it for speed.
  • Use categorical output types when you know the likely answers in advance.
  • Enable Allow multiple values when a document can contain more than one valid answer.
  • Save question templates for recurring tasks and share them within a workspace.

Known Limitations​

  • Answers vary by document quality and indexing. Documents without indexed content are skipped and flagged.
  • Results depend on question phrasing. Vague questions produce inconsistent or incomplete answers.
  • Multiple-answer cases require configuration. If you do not enable Allow multiple values, results may be truncated to a single value.