Virbe Documentation
  • Say hi to Virtual Beings!
  • Getting started
    • Introduction to Virbe platform
      • Virbe-hosted launch
      • Azure-hosted deployment
  • Dashboard Management
    • Dashboard Architecture
    • Overview
    • Profiles
      • General
      • Language
      • Persona
      • Settings
        • Web widget profile settings
        • Metahuman kiosk profile settings
      • Deployment
        • Web widget deployment
        • Metahuman kiosk deployment
    • Configurations
      • Speech-to-Text
      • Text-to-Speech
      • AI Models
      • Conversational Engines
    • Personas
    • Knowledge Base
      • Best Practices for Knowledge Base Content
    • Data tables
    • Conversation Flows
      • Getting started with Conversation Flows
      • Managing nodes
      • Nodes
        • Logic nodes
          • Route by Profile
          • If/Else Router
          • Intent Matcher
        • Flow control nodes
          • Go to Flow
          • Checkpoint
        • Response nodes
          • Text
          • LLM Response
          • Call ConvAI
          • Behavior
        • Action nodes
          • Collect User Data
          • Quick Reply
          • Cards from Table
        • Context nodes
          • Find Record(s)
        • Integration nodes
          • Call Webhook
          • Custom Action
      • Troubleshooting Conversation Flows
  • Touchpoints
    • Kiosk Apps
      • Metahuman Kiosk
        • Hardware setup
          • Recommended setup
          • Computer
          • Screen
          • Microphones
          • IoT
          • Peripherals
          • Operating System
            • Windows
              • Login & power management
              • Other
              • OpenHab
            • Mac
            • Remote Access
        • Customer Experience
          • VAD & press-to-talk
          • Microphone settings
          • Avatars
    • Web Integration
Powered by GitBook
On this page

Was this helpful?

  1. Dashboard Management
  2. Knowledge Base

Best Practices for Knowledge Base Content

Content Organization for LLM Processing

Document Length and Chunking

  • Keep individual documents focused and concise rather than creating long, comprehensive documents

  • Aim for natural breakpoints in content - each document should cover one complete concept or topic

  • Consider how information might be retrieved - break up content based on likely user queries

  • The 10,000 character limit isn't a target - shorter, focused documents often work better

Information Redundancy

When similar information appears in multiple documents:

  • Redundancy can confuse LLM context understanding

  • Best approach: Reference a single source document for core information, then add context-specific details in other documents

Example: ✅ Good:

  • Core document: "Product X Technical Specifications"

  • Related docs: "Product X for Beginners", "Product X Troubleshooting" Each adds unique context while referencing core specs

❌ Avoid:

  • Multiple documents repeating the same specifications with slight variations

  • Inconsistent versions of the same information

Handling Conflicting Information

  • Maintain a single source of truth for factual information

  • If information changes over time, update all related documents

  • For genuinely conflicting scenarios (e.g., different recommendations for different situations), clearly specify the context

Content Structure

Document name

The way you name the document should serve the purpose of intuitive document organization, however to make sure that it is included in the relevant chunks, repeat it at the top fo the document content.

Information Hierarchy

  • Start with the most important information

  • Use clear headings and logical grouping

  • Build from general to specific

  • Include relevant context without overloading

Context Signaling

Help the LLM understand content relationships:

  • Use clear transitional phrases

  • Explicitly state relationships between concepts

  • Include relevant qualifiers and conditions

Example: "This guidance applies specifically to Model X-100 manufactured after 2024."

Language and Style

Clarity and Consistency

  • Use consistent terminology throughout documents

  • Define technical terms when first used

  • Maintain consistent formatting for similar types of information

  • Use clear, unambiguous language

Contextual Markers

Include phrases that help LLMs understand:

  • Purpose: "This document explains..."

  • Scope: "This applies to..."

  • Relationships: "This is related to..."

  • Conditions: "Only valid when..."

Optimizing for Retrieval

Keywords and Natural Language

  • Include natural variations of key terms

  • Use complete sentences rather than bullet points

  • Incorporate likely user query phrasing

  • Balance technical accuracy with conversational tone

Common Pitfalls to Avoid

Content Issues

  • Overly complex sentences that may confuse context

  • Ambiguous pronouns or references

  • Implicit knowledge not stated in the text

  • Inconsistent terminology

Structure Issues

  • Too much information in one document

  • Poor organization that obscures relationships

  • Lack of clear context or scope

  • Missing crucial qualifiers or conditions

PreviousKnowledge BaseNextData tables

Last updated 3 months ago

Was this helpful?