Articles in this section

Knowledge Sources in Presentation Generation

Updated

This article describes how uploaded files and preconfigured knowledge contribute to the generation of presentations by the Universal and Custom Agents.

  Prerequisites

  • Library and Universal/Custom Document Agent modules enabled.
  • Space owner access to the Templafy tenant.

Overview

Knowledge files are documents that contain information the AI uses to generate accurate and contextually relevant content. Instead of relying solely on general knowledge, the AI can access specific organizational data, guidelines, product information, or any other relevant content stored in these files. There are two types of knowledge that the system uses:

  1. User-Uploaded Files (Chat Files)
  2. Knowledge Collections (Preconfigured Knowledge)

  Note

Both knowledge types can be used with Universal and Custom Document Agents.

User-Uploaded Files (Chat Files)

Files that users upload during their conversation with the Document Agent.

Characteristics

  • Uploaded directly in the chat interface.
  • Temporary and tied to a specific session.
  • Processed and indexed when uploaded.
  • Available only for the current presentation being created.

Processing Details

  1. User uploads a file during the chat.
  2. The file is stored securely.
  3. The content is extracted and broken into smaller chunks.
  4. Each chunk is analyzed and indexed for semantic search.
  5. The file becomes searchable within seconds to minutes.

Supported Formats

  • PDF Documents.
  • Word Documents (.docx).
  • PowerPoint Presentations (.pptx).
  • Text Files.

Size Limits

  • Maximum file size: 50 MB.
  • Maximum content size: 500,000 characters.

Knowledge Collections (Preconfigured Knowledge)

Files set up by administrators as part of Universal and Custom Agents.

Characteristics

  • Configured by tenant administrators.
  • Persistent and reusable across sessions.
  • Can include usage instructions for the AI.

Types of Knowledge

Type Description Source
Files Documents uploaded to the knowledge collection Internal file storage
Copilot Retrieval Integration with Microsoft 365 Copilot Microsoft Graph API
Glean Integration with Glean enterprise search Glean API

Configuration Options

  • Maximum number of results to retrieve (default varies by type).
  • Usage instructions telling the AI how to use this knowledge.

Knowledge in the Chat Phase

During the chat phase, the AI gathers information about what the user wants to create. Knowledge files help the AI understand context and answer questions accurately.

  Note

  • Agents are instructed to always check uploaded files before asking questions.
  • Agents verify information exists in knowledge sources before including it.
  • Agents do not invent or assume facts that aren't in the knowledge sources

User-Uploaded Files (Chat Files)

When a user uploads a file, the AI:

  • Searches through the uploaded files to find relevant passages using semantic search (meaning-based, not just keyword matching).
  • Extracts and summarize relevant information.

Example

  • User: "Create a presentation about our new product launch"

  • User: [Uploads ProductLaunch_Plan.pdf]

     

  • AI: [Searches the uploaded file for product information]
  • AI: "I found information about the Q2 product launch in your document. 
  • AI: I see it covers the new features, target market, and pricing.
  • AI: Would you like me to include all these topics in the presentation?"

Knowledge Collections (Preconfigured Knowledge)

When using a Custom Agent with knowledge collections, the AI:

  • Queries multiple knowledge sources.
  • Retrieves information relevant to the presentation topic.
  • Combines information from multiple sources.

Example

  • User: selects "Quarterly Business Review" Custom Agent
    (This agent has knowledge collections for company financials and goals)

  • User: "Create a QBR presentation for Q4"

     

  • AI: [Searches the uploaded file for product information]
  • AI: "I found information about the Q2 product launch in your document. 
  • AI: I see it covers the new features, target market, and pricing.
  • AI: Would you like me to include all these topics in the presentation?"

Knowledge in the Outline Phase

The outline phase determines the structure of the presentation - which slides to create and what topics each slide will cover.

User-Uploaded Files (Chat Files)

When a user uploads a file, the AI:

  • Verifies information exists before adding it to the outline.
  • Avoids creating slides that cannot be backed by actual data.

Knowledge Collections (Preconfigured Knowledge)

When using a Custom Agent with knowledge collections, the AI:

  • Works within the constraints of preconfigured outline templates, including having mandatory and recommended sections.
  • Ensures comprehensive coverage of required topics using multiple knowledge sources.

  Note

Before finalizing the outline, the system:

  • Confirms each proposed slide has supporting information.
  • Notes any gaps where information might be missing.
  • Suggests alternatives if requested topics can't be supported.

Related to

admin_role
Was this article helpful?
0 out of 0 found this helpful

Comments

0 comments

Article is closed for comments.