The Data Analyst Assistant is an artificial intelligence assistant in GeneXus Enterprise AI that allows the end user to interact with a dataset in a similar way as a professional data analyst would.
This assistant offers functions such as data summaries, calculations of maximum, minimum and average values, data organization in tables and report generation.
To define an assistant of this type, you must provide the following information:
- Metadata: A description of each dataset, its columns (including data types and possible values), and considerations for the coder and the interpreter (see Define Metadata (JSON)).
- Glossary: A list of terms used within the company and/or the domain of the user that would be relevant for the LLM to understand the questions.
- Datasets: A set of CSV files that follows the descriptions added as metadata.
When an end user interacts with the Data Analyst Assistant through the Frontend, GeneXus Enterprise AI performs a series of tasks to identify the purpose of the question, select the appropriate dataset, and generate code to extract the relevant information and present it in a friendly way.
This process involves several key components:
- Condenser: This component reformulates the end user's input into a prompt that is easier for the underlying LLM to understand and process.
- Classifier: The classifier determines the appropriate data source to use based on the user's question.
- Coder: Generates the necessary code to query the selected data source for the relevant information. It uses Metadata to guide this process of extracting relevant information from the Datasets.
- Interpreter: The interpreter runs the code generated by the Coder, retrieves the information, and generates the final answer, integrating the retrieved data in a user-friendly format.
To facilitate the process of extracting the correct information, the Data Analyst Assistant includes a Conversational Module that allows the end user to interact in a fluid, conversational way.
The Conversational Module uses the contents of the Glossary and Metadata solely to understand the types of questions it can answer and the data it contains.
The key components use the Glossary to refine the question, Metadata to generate the code that queries the Datasets, and the Datasets to retrieve the relevant information needed to respond.
For example, before requesting an analysis of the data, the user can query terms from the Glossary, the assistant's capabilities, or the data structure defined in the Metadata.
For these kinds of questions, the Conversational Module directly retrieves the Metadata and Glossary definitions from the assistant, bypassing the data query steps. This ensures that users receive fast and accurate answers for Glossary terms without the need for additional data processing.
To define or update a Data Analyst Assistant, you must use the GeneXus Enterprise AI Backoffice and follow the steps outlined in How to use Data Analyst Assistant.