Skip to content

RAG_DOCUMENTATION.md

Latest commit

 

History

History
134 lines (97 loc) · 4.52 KB

RAG_DOCUMENTATION.md

File metadata and controls

134 lines (97 loc) · 4.52 KB
Raw

RAG (Retrieval-Augmented Generation) Feature Documentation

Overview

The RAG feature enhances the mod-ollama-chat module by providing bots with relevant contextual information from a knowledge base when responding to player messages. This allows bots to give more accurate, detailed, and informative responses about World of Warcraft game mechanics, professions, classes, and other topics.

How It Works

  1. Knowledge Base: JSON files containing structured information about WoW topics
  2. Query Analysis: When a player sends a message, the system analyzes it for relevant keywords
  3. Information Retrieval: The system searches the knowledge base for related information
  4. Prompt Enhancement: Retrieved information is added to the bot's prompt before generating a response
  5. Contextual Responses: Bots can now provide detailed information about professions, classes, etc.

Configuration

Basic Settings

# Enable/disable the RAG feature
OllamaChat.EnableRAG = 1

# Path to RAG data files (relative to module data directory)
OllamaChat.RAGDataPath = rag/

# Maximum number of information items to retrieve
OllamaChat.RAGMaxRetrievedItems = 3

# Minimum similarity score for information to be considered relevant (0.0-1.0)
OllamaChat.RAGSimilarityThreshold = 0.3

# Template for including RAG information in prompts
OllamaChat.RAGPromptTemplate = "RELEVANT INFORMATION:\n{rag_info}\nUse this information to provide accurate and detailed responses when applicable."

Data Format

RAG data is stored in JSON files in the data/rag/ directory. Each file contains an array of entries with the following structure:

[
  {
    "id": "unique_identifier",
    "title": "Display Title",
    "content": "Detailed information about the topic",
    "keywords": ["keyword1", "keyword2", "keyword3"],
    "tags": ["tag1", "tag2"]
  }
]

Field Descriptions

  • id: Unique identifier for the entry (used internally)
  • title: Human-readable title for the information
  • content: The detailed information that will be provided to the bot
  • keywords: Array of keywords that help match queries to this information
  • tags: Array of category tags for organization

Example Usage

Player Message: "How do I become an alchemist?"

Without RAG: Bot gives a generic response about professions With RAG: Bot provides detailed information about Alchemy, required skills, training locations, and key recipes

Player Message: "What's the best class for raiding?"

Without RAG: Generic class advice With RAG: Detailed information about different classes, their roles, strengths, and raid viability

Creating Custom Knowledge Base

  1. Create JSON files in data/rag/ directory
  2. Follow the JSON structure above
  3. Use relevant keywords that players might use in queries
  4. Keep content concise but informative
  5. Test with various query patterns

Tips for Good Keywords

  • Include common misspellings
  • Use both singular and plural forms
  • Include related terms (e.g., "potions" for alchemy)
  • Consider how players actually phrase questions

Performance Considerations

  • Memory Usage: All RAG data is loaded into memory on startup
  • Query Speed: Retrieval uses simple text similarity, very fast
  • Token Limits: Retrieved information adds to prompt length
  • Relevance Filtering: Similarity threshold prevents irrelevant information

Troubleshooting

RAG Not Working

  1. Check that OllamaChat.EnableRAG = 1
  2. Verify JSON files exist in data/rag/ directory
  3. Check server logs for RAG initialization messages
  4. Ensure JSON syntax is valid

Poor Relevance

  1. Adjust OllamaChat.RAGSimilarityThreshold (lower = more results, higher = fewer but more relevant)
  2. Add more keywords to your JSON entries
  3. Review query preprocessing logic

Performance Issues

  1. Reduce OllamaChat.RAGMaxRetrievedItems
  2. Increase OllamaChat.RAGSimilarityThreshold
  3. Optimize JSON file sizes

Examples

Professions Data

See wow_professions.json for examples of profession information including:

  • Alchemy (potions, elixirs, flasks)
  • Blacksmithing (weapons, armor, tools)
  • Enchanting (disenchanting, enchants)
  • And more...

Classes and Factions Data

See wow_classes_factions.json for examples of:

  • Class descriptions and roles
  • Faction information and races
  • Specializations and abilities

Future Enhancements

  • Support for more advanced similarity algorithms
  • Dynamic knowledge base updates
  • Integration with external knowledge sources
  • Multi-language support
  • Category-based filtering