Skip to content

RAG

Key Concepts

The Knowledge Base Updater block is a specialized block built on Retrieval-Augmented Generation (RAG). It is designed to retrieve highly specific information from uploaded documents and provide precise answers by minimizing the risk of LLM hallucination.

This block combines traditional information retrieval techniques like BM25 with neural ranking methods such as ColBERT (when available), making it ideal for use cases where the accuracy and contextual relevance of responses are critical.

A common use case includes defining a business process from a document (e.g., PDF), then using that process to guide further actions such as database queries or workflow decisions.

Key Definitions

Term Definition
RAG (Retrieval-Augmented Generation) A technique that enhances LLMs by retrieving relevant data from a knowledge base before generating responses.
BM25 A ranking function used in traditional keyword-based retrieval systems.
ColBERT A neural retrieval model that improves search quality by comparing token-level embeddings.
Knowledge Base A collection of structured or unstructured files used for document-level retrieval in workflows.
Workflow Node A processing unit within a workflow that performs a specific task (e.g., retrieve, rerank, synthesize).

Setup Guide: Configuring a Knowledge Base

To create and integrate a Knowledge Base into your INTELLITHING workflow, complete the following setup steps:

1. Knowledge Base Name

  • Enter a unique and descriptive name for your knowledge base.
    Example: Customer Support Knowledge Base, Product FAQ Repository

2. Knowledge Base File

  • Upload one or more documents containing the information you want to use.
  • Supported file formats: .txt, .json, .pdf
  • Click Add to Knowledge Base Files List to confirm and upload the files.

3. Description

  • Provide a short description of the knowledge base contents and its intended purpose.
    Example: A repository of FAQs and troubleshooting guides for customer service.

  • While optional in custom workflows, it helps the routing engine or other team members understand the knowledge base’s context.

4. Save

  • Click Save to finalize your configuration and return to the main editor.

Workflow Nodes

The Knowledge Base Updater includes a set of RAG-powered workflow nodes:

rag_retrieve_nodes (First Node)

  • Retrieves the most relevant information from the knowledge base based on the query or context.
  • Ensures the workflow is grounded in authoritative, document-sourced knowledge.

rag_rerank_nodes

  • Reranks the retrieved documents or excerpts based on their relevance to the query or workflow intent.
  • Surfaces the most contextually appropriate content.

rag_synthesize (Last Node)

  • Synthesizes the selected and reranked information into a coherent summary, answer, or action recommendation.
  • Prepares the information for use in subsequent workflow steps or user responses.

Best Practices for Knowledge Base Management

  • Use Clear Names: Name your knowledge bases in a way that reflects their content and purpose.
  • Organize Files Thoughtfully: Ensure uploaded files are clean, relevant, and well-formatted.
  • Add Descriptions: Descriptions help route queries accurately and simplify collaboration with other users or teams.

Example Use Case

You want to automate internal support workflows using an operations manual PDF:

  1. Upload ops_manual.pdf to a new knowledge base named Internal Operations Manual.
  2. When a workflow receives a query like "How do I escalate a P1 issue?", the rag_retrieve_nodes fetches relevant sections from the manual.
  3. rag_rerank_nodes filters the content for the most useful parts.
  4. rag_synthesize produces a clean, readable summary of the escalation steps.