Requirements Ingestion Agent - User Guide

The Requirements Ingestion Agent is a Task Assistant Agent within the InnovatorEdge AI framework.
It enriches the digital thread by transforming requirements contained in unstructured PDF documents into structured, governed Items within Aras Innovator Requirements Engineering (RE).

The agent operates directly inside the system of record to normalize incoming document data. Its purpose is to:

• Extract requirement statements from PDF documents.
• Separate engineering content from document presentation and formatting.
• Convert extracted content into governed Requirement Items under a Requirements Document Item.
• Reduce manual transcription effort for large technical specifications.

The quality of extraction depends on the structural conventions of the source PDF file.
The agent analyzes structural patterns, including:

• Requirement Boundaries: Whether requirements are clearly separated by numbering and/or headings.
• Directive Phrases: Presence of action-oriented statements (e.g., “Add,” “Replace”) indicating modification actions.
• Integrated Content: Identifying whether lists, tables, or images are part of requirement statements.
• Document Commentary: Distinguishing between core requirements and notes, justifications, or general commentary.

Constraints:

• Format: Supports a single .pdf file.
• Size: Maximum file size is 10 MB.
  

Agent Instructions configure how the AI interprets the document during the ingestion phase.
These apply globally across the document and shape how boundaries and titles are determined.

Examples:

• Section-Based Rules: “Start a new requirement at each hierarchical section number such as 6.1.1. Use the section number as the Requirement Title.”
• Action-Based Rules: “Create a separate requirement for each action sentence that begins with Add, Replace, or Delete.”
• Exclusion Logic: Instructions can specify what content to ignore, such as “Do not treat Notes as requirements”.

During parsing and segmentation, the agent identifies and segments the following content types:

• Text: Paragraph-based statements extracted as structured text, which is fully editable during review.
• Tables: Tabular content associated with a requirement is preserved. Table content is not editable during review but can be adjusted later in Aras Innovator.
• Graphics: Images and diagrams are identified and prepared for import as Requirement Item content.

Once the import is initiated, the agent enters a multi-stage technical extraction process:

1. Parsing: Analyzing structural elements like headings, numbering schemes, and paragraph blocks.
2. Content Detection: Identifying specific segments that contain requirement-relevant data.
3. Formalization: Constructing candidate Requirement entities containing derived titles, text, and related element references.
   

If the agent encounters complex formatting or connection issues it cannot resolve, it will report an Import failed status.
In the event of a failure, the specific cause is displayed (e.g., “Failed to classify page n”).
Selecting Try again returns you to the Upload screen to attempt ingestion again with adjusted instructions.

After processing is complete, the agent presents the Review Requirements interface. This phase allows engineers to apply domain expertise to the AI’s interpretation, ensuring the structured model is accurate before it is committed to Aras Innovator.

The interface uses a split-view layout designed for high-volume auditing:

• Navigation Pane (Left): A vertical, scrollable list of all extracted items.
• Persistent Search: A Find bar remains docked at the top of the navigation pane, allowing you to filter items by keyword even while scrolling through long documents.
• Element Details (Right): A contextual pane that displays the full content of the item currently selected in the navigation list.

Each extracted item in the list is explicitly labeled with its identified type and a corresponding icon to help users quickly distinguish between different content blocks:

• Requirement (Bullseye icon): Formal engineering requirements.
• Text (Three-line icon): Standard paragraph text or section headers.
• Table (Grid icon): Data structured in rows and columns.
• Graphic (Picture icon): Visual elements, which the agent identifies as either photos/pictures or UML Diagrams.
  

The interface provides several tools for refining the extracted data model:

• Drag-and-Drop Reorganization: Each item in the navigation pane features a handle icon on the far left, allowing you to drag and drop items to reorder them or change the document hierarchy.
• Inline Editing: Selecting a Requirement or Text item opens an editable view in the Element Details pane. You can manually update Requirement Titles and body text to correct any AI parsing errors.
• Adding Content: Within the details pane for Text elements, you can select + Title or + Text to add new headers or paragraphs that may have been missing from the initial extraction.
• Deleting Elements: Any irrelevant content—such as headers, footer text, or discarded notes—can be permanently removed by selecting the Delete option at the bottom of the details pane.
• Inspecting Graphics and Tables:
  
  • Graphics: You can view high-resolution previews of extracted images and diagrams to ensure they are associated with the correct requirement.
  • Tables: The details pane renders the extracted grid for visual verification. While table content is not editable in this view, the structure is preserved for the final import.

Once validation is complete, you provide a Requirement Document Title and initiate the formal import sequence.

The agent creates the following native objects:

1. Requirements Document Item: Exactly one parent container is created per session.
2. Requirement Items: Individual items are generated for every validated extraction.
3. Relationship Binding: Structured links connect the document to its requirements in the correct sequence.
4. Graphic Commitment: Images and diagrams are uploaded as File Items and associated with the corresponding records.

Upon successful completion, an Import Summary is displayed.

Selecting Open in Innovator opens the standard interface, where the document and requirements participate in standard governance, including lifecycle management and versioning.

• Human-in-the-Loop: All AI-generated extractions must be reviewed and confirmed by an engineer before system commitment.
• Structural Consistency: Extraction quality is highest when documents follow consistent section and numbering conventions.
• Permission Awareness: The import process respects your native Aras Innovator permissions and assigned roles.
• Error Handling: If an ingestion job fails (e.g., a classification failure), the system provides a specific error message and a Try again option to restart the process.