Chat bot Text-to-SQL using Azure AI Services

Question

Hey, im currently working on a chat bot that based on a user question it generates a query to the database and then it processes the result of that query in natural language. Im currently using Azure OpenAI and Azure AI Search to index the data source that will be used. 1.Is this possible? 2.What structure should the file that will be used by AI Search have, so that it contains information about the tables, columns and the possible distinct values for each column and also prevent necessary information to be lost when chunking. Considerations: The tables have a "Total" column that represents the amount of "X" that are defined by the combination of all the other column values. The average row count for each table is 800 000.

Answer 1

Hi there oão Conceição

Thanks for using QandA platform

Yes, it’s possible to build a chatbot that generates SQL queries from natural language inputs and then processes the query results in natural language using Azure OpenAI and Azure AI Search. to do this, the file used by Azure AI Search should be structured to include metadata about your database schema. You could structure the data to include the table name and a list of columns along with their distinct values. For chunking, you index the schema metadata separately from the data rows to avoid losing critical information, especially for large tables. The "Total" column representing aggregated values can be included in the index as well to help with generating queries involving sums

If this helps kindly accept the answer thanks much.