VisualsPDF version

AI visual

Cloudera Data Visualization enables you to embed interactive, intelligent components into your dashboards and applications. The AI Visual feature brings AI to your BI workflows. With the power of retrieval-augmented generation (RAG) inside the dashboard, you can explore AI-driven insights alongside traditional analytics, query large datasets in natural language, and pass the context of a conversation to structured reports for dynamic decision-making.

The AI visual provides a natural language interface for interacting with datasets through text or speech. It combines conversational input with vector-based data search and large language model (LLM) capabilities to generate insightful, contextual responses, making it a powerful tool for enhancing your dashboards.

You can configure an AI visual using Cloudera Data Visualization’s drag-and-drop dashboard builder. Before publishing, you can customize the welcome message shown on the dashboard, the prompt that is sent to the AI model, and token limits for each response. These settings enable you to manage response length, optimize performance, and control costs as well as provide your business users with an overview of how to leverage this powerful AI analytics solution.

Interacting with the AI visual

Once the AI visual is configured and added to your dashboard, you can begin exploring your data through conversational queries. You can ask questions about your data using voice or text. The visual generates responses based on the connected dataset, making it easy to gain insights using natural language.

  • Text input: Type your question into the input field. The AI visual processes your query and returns a relevant response based on the dataset.

  • Speech input: Click the to ask a question using your voice. Your speech is converted to text and entered into the input field. Click SEND to submit the query.

  • Follow-up queries: Continue the conversation by asking follow-up questions or refining your query using either voice or text input.

Exploring the data behind the responses of the AI visual

  • To view the source data used for a response, click the next to the reply.

  • The displayed information depends on what's configured in the Tooltip shelf for the AI visual.

  • You can explore other dashboard components to validate insights, compare data points, or perform further analysis.

  • The AI visual can also apply smart filters to your dashboard based on your questions. For instance, asking about a specific topic will automatically adjust other visuals to highlight relevant data. This helps validate the AI’s response and offers users a more intuitive way to explore insights, without the need to manually search through filters or menus.

When added to a dashboard, the AI visual connects to a vector database (Solr 9 or SQLite). Based on user input (typed or spoken), it performs a similarity search on the vectorized data. The matching records are sent to a large language model (configured in Site Settings), which generates a contextual response shown within the visual.

To use the AI visual, you must connect to a data source that contains a vector column with pre-created embeddings. This can be done in different ways depending on how your data is stored and ingested.

One common method is to upload a CSV file into a supported vector database (Solr or SQLite). For instructions, see Importing data in CSV format.

Another option is to use continuous data ingestion. If your dataset is updated continuously (for example stored in an Apache Iceberg table), you can use a data pipeline to:

  • Convert new records into vector embeddings
  • Write them into the connected vector database

This approach allows you to keep the dataset used by the AI visual up to date and relevant with minimal manual effort.

You can configure an AI visual using Cloudera Data Visualization’s drag-and-drop dashboard builder. To create an AI visual, follow the step-by-step instructions below. For details on configuring its shelves, see Shelves for AI visual.

Before creating the AI visual, make sure the following prerequisites are met.

Configuration:

Data requirements:

  • Ensure the data you plan to use with the AI visual is vectorized and available in a vector database.
  • Connect to a supported vector database.
  1. Connect to a vector database that contains your dataset with vector embeddings.
    This can be from a CSV upload or a continuously updated data source.
  2. Create a dataset from the vectorized data.
    For instructions, see Creating a dataset.
  3. Create a new dashboard.
    For instructions, see Creating a dashboard.
  4. Add the AI visual to the dashboard.
    1. In the VISUALS menu, click the icon of the AI visual.
    2. Populate the visual shelves from the fields in the DATA pane.
      1. Embeddings: Add vector fields containing embeddings.
      2. Embedding Context: Add fields that you want to include in the chat prompt.
      3. Context Aggregates: Add aggregated fields that summarize data for similar records in the AI visual. This allows the AI sisual to better answer questions related to quantities, distributions, and comparisons.
      4. Tooltip: Define the source information to be included in the result. This information appears when the user hovers over the response's Info icon, but it is not sent to the completion service.
      5. Limit: Define the maximum number of data rows from the vector database that are processed by the visual.

      For more information, see Shelves for AI visual.

  5. Click Settings in the right-side VISUAL menu to configure the visual settings.
    1. Display
      1. Customize the chat welcome message, which is displayed as the first chat message from the AI visual.
      2. Set the display format for the conversation: plain text or markdown
    2. Embeddings
      1. Set the maximum number of tokens. Use 0 or a negative value to disable this setting.
      2. Select or change the default profile.
    3. Completion
      1. Select or change the default profile.
      2. Set the maximum number of tokens. If exceeded, the request is not sent to the service and an error message is displayed.
      3. Choose a context overflow policy that defines how to handle situations where the length of generated tokens exceeds the context window size:
        • Throw an error: Raises an error when the conversation length exceeds the defined context window.
        • Remove old conversation first: Automatically manages conversation length by removing old tokens.
        • Truncate embedding context first: Truncates the embedding context, retaining the latest tokens.
      4. Specify the context prompt (first prompt message) for the chat.
      5. Specify the question prompt, which is the formatting of the user question.

We want your opinion

How can we improve this page?

What kind of feedback do you have?