Data Visualization

An llm does not have the innate capability to make dynamic visualizations like charts and KPI cards.

Therefore, I manually make an agent to make appropriate visualization from the final response.

def visualization(llm, answer):
    template = """
    You are given a statement which may contain one or multiple quantities. You must return 'compare', 'time-series' or 'kpi' depending on whether the statement contains data comparing similar quantities of multiple items, or a time series of data, or a single quantity, as well as the corresponding data and
    categories if any, and type of quantity like quantity delivered or inventory.

    ---
    <Instructions>
    ---

    ---
    <Examples>
    ---

    ---
    Statement: {answer}
    Response:
    ---
    """

    prompt_template = PromptTemplate(input_variables=["answer"], template=template)
    return llm.invoke(prompt_template.format(answer=answer)).content

The output contains information of what visualization to use and data to be visualized, so it is passed directly to the frontend where an appropriate component is rendered conditionally.