Skip to main content

Knowledge Base

A knowledge base (KB) is a structured repository of information designed to store and organize data that a retrieval model can access to find relevant context for a language model (LLM) to generate accurate responses. The KB in RAG systems acts as the "memory" from which relevant knowledge is retrieved and used to enhance response quality.

Key Components and Functionality of a Knowledge Base in RAG​

Data Sources and Parsing​

The KB typically consists of various types of data, including documents (FAQs, guides, reports), structured datasets, and other relevant content. Data is preprocessed and parsed into smaller chunks (paragraphs, sentences, or user-defined sections) to enable efficient retrieval.

Embedding and Vectorization​

Each chunk of text is converted into an embedding (a numerical vector that captures the semantic meaning of the text) using models such as BERT, RoBERTa, or specialized embedding models. These embeddings allow the KB to organize data in a way that enables similarity-based search, enabling the retriever to locate contextually relevant information efficiently.

The KB is indexed to support rapid retrieval based on vector similarity. This indexing process enables matching user queries with relevant chunks in the KB based on semantic similarity. Indexing methods such as FAISS (Facebook AI Similarity Search) are commonly used to optimize this process.

Retrieval of Contextual Data​

When a user query is received, the RAG system’s retriever searches the KB for chunks that closely match the meaning of the query. The retrieved chunks are then passed to the LLM, which uses them as context to generate a more accurate and informed response.

Knowledge Base Updating and Maintenance​

The KB in a RAG system can be updated with new information as it becomes available, ensuring that the system remains accurate and up to date. Changes in company policies, new product information, and other relevant updates are incorporated into the KB, and embeddings are recalculated as required.

Getting Started​

Create a New Knowledge Base​

Step 1: Log in to the TIR AI Platform​

Ensure that you are logged in and working within the correct project. If required, create a new project before proceeding.

Step 2: Navigate to the Knowledge Base Section​

From the TIR dashboard:

  • Locate the primary navigation panel (typically on the left side of the interface)
  • Select the RAG or Knowledge Base section
  • This will open the page where all existing knowledge bases are listed

Step 3: Create a Knowledge Base​

On the Knowledge Base listing page:

  • Locate the Create Knowledge Base button (typically positioned in the top-right corner)
  • Click this button to begin the creation process

Step 4: Configure Knowledge Base Details​

You will be redirected to the Create Knowledge Base page.

On this page:

  • Fill in all required fields presented in the form
  • Ensure all mandatory configurations are completed before proceeding

Model Source​

Users can choose the model source as either GenAI or Endpoint.

  • If GenAI is selected:

    • Open the embedding model dropdown
    • Select a predefined embedding model from the available options

  • If Endpoint is selected:

    • Select the embedding model from the model endpoint list
    • Select the corresponding model name
    • Perform model validation to ensure that the selected model exists and is available for use
    • Only proceed after successful validation

Embedding Model​

Embedding models in a knowledge base are used to convert textual data (or other data types) into numerical representations, known as embeddings, that capture the semantic meaning of the content. These embeddings enable the knowledge base to perform efficient search, comparison, and retrieval of relevant information based on meaning, rather than relying solely on exact keyword matching.

Model Source (Repeated Configuration)​

Users can again verify or modify the model source configuration as either GenAI or Endpoint.

  • If GenAI is selected, ensure a predefined embedding model is selected.
  • If Endpoint is selected, ensure both embedding model and model name are selected and validated before proceeding.

Chunking Method​

Chunking in a knowledge base is a method of dividing larger documents or datasets into smaller, manageable units of information, referred to as "chunks." This approach is particularly important in systems that rely on retrieval-augmented generation (RAG) models or semantic search techniques, as it improves the system’s ability to retrieve relevant context and generate accurate responses by operating on smaller, more focused pieces of content.

Chunk Size​

Chunk size in a knowledge base refers to the length of each individual chunk, which may be defined in terms of words, tokens, sentences, or characters. Selecting an optimal chunk size is critical, as it directly impacts context preservation, retrieval accuracy, and computational efficiency. Proper tuning of chunk size ensures that each chunk contains sufficient contextual information while maintaining efficient processing and retrieval performance.

Delimiter​

A delimiter in a knowledge base is a character or sequence of characters used to separate sections of text or data. Delimiters play a key role in chunking, parsing, and structuring content within the knowledge base, enabling the system to accurately process, segment, and retrieve information in a consistent and reliable manner.

Questions​

Automatically extract N questions for each chunk to increase their ranking for queries containing those questions.

  • After chunk creation, navigate to the chunk list within the knowledge base
  • Review the automatically generated questions for each chunk
  • Modify or update the questions if required

Additional considerations:

  • If an error occurs during question extraction, the chunking process will continue
  • In such cases, empty results may be added to the original chunk
  • This feature consumes additional tokens from the LLM configured in the system model settings

Step 5: Launch Knowledge Base​

After completing all configurations:

  • Locate the Launch button on the Create Knowledge Base page
  • Click Launch to finalize and create the Knowledge Base

Knowledge Base List​

After creation, all knowledge bases are displayed in a list view.

  • Each row represents an individual knowledge base
  • To manage a knowledge base:
    • Click on the corresponding row or entry
    • This will open the detailed view for that knowledge base

Documents​

After selecting a knowledge base, you will be redirected to the Documents section.

Dashboard​

The dashboard displays:

  • A table of all uploaded documents
  • Document status
  • Available actions for each document

Import Data​

To upload data into the knowledge base:

  • Locate and click the Import Data button within the Documents section
  • This opens the data import interface

You can upload data using two methods:

Browse Files​

  1. In the import interface, locate the upload area labeled: “Drop a file or click to upload”
  2. Click the upload area or drag and drop files
  3. Select one or more files from your local system
  4. After selection, click the Import button to upload the files

Import from TIR Dataset​

  1. In the import interface, locate the TIR Dataset dropdown
  2. Click the dropdown to view available datasets associated with your account
  3. Select the desired dataset
  4. Choose the specific file within the dataset
  5. Click the Import button to add the file to the knowledge base

Bulk Actions​

Bulk actions allow operations to be performed on multiple documents simultaneously.

  • In the document table, use the selection checkboxes to select multiple documents
  • Once selected, use the bulk action controls located above the table
  • Apply the desired action (such as parsing or deletion) to all selected documents

Status Filter​

The status filter allows you to quickly locate documents based on their processing status.

  • Locate the filter control above the document table
  • Select a status (e.g., processed, pending, failed)
  • The table will update to display only documents matching the selected status

Actions​

Each document entry in the table includes the following actions:

Parsing​

  • Locate the parsing option within the document row
  • Click the parsing action to initiate processing for the selected document
  • This is required for documents that have been newly uploaded or modified

Edit​

  • Locate and click the Edit option for a document
  • A configuration panel will open
  • Update parameters such as:
    • Chunking method
    • Chunk size
    • Delimiter
  • Save changes to apply the updated configuration

Delete​

  • Locate the Delete option within the document row
  • Click Delete to remove the document
  • Confirm the action when prompted

Download​

  • Locate the Download option within the document row
  • Click Download to save the document to your local system

Retrieval Testing​

Retrieval testing evaluates how effectively the system retrieves relevant information based on user queries.

Key Components of Retrieval Testing​

Test Queries​

  • Develop a set of diverse test queries representing real user inputs
  • Include:
    • Keyword-based queries
    • Natural language questions
    • Entity-specific queries

Expected Results​

  • Define expected results for each test query
  • Use these results to evaluate the relevance and correctness of retrieved data

Retrieval Mechanism​

  • Identify the retrieval approach used by the system:
    • Keyword matching
    • Semantic search
    • Embedding-based retrieval
  • Ensure that the retrieval mechanism aligns with the intended system architecture

Performance Metrics​

Measure retrieval effectiveness using:

  • Precision: Ratio of relevant results retrieved to total results retrieved
  • Recall: Ratio of relevant results retrieved to total relevant results available
  • F1 Score: Harmonic mean of precision and recall
  • Response Time: Time taken to retrieve and present results

Configuration​

Within a specific knowledge base, the Configuration section allows updates to:

  • Chunk size
  • Delimiter
  • Questions

To update configuration:

  • Navigate to the Configuration section of the selected knowledge base
  • Modify the required parameters
  • Save changes to apply updates

Data Imports​

The Data Imports section allows you to manage data synchronization and view logs.

To access:

  • Navigate to the Data Imports tab within the knowledge base

Within this section:

  • Initiate data synchronization processes
  • View logs, including timestamps and status of import operations