sql-query-generator

SQL Query Generator & Executor Example

This demonstrates a simple application using public Northwind with Amazon Titan Embeddings, Amazon Bedrock Claude Model, LangChain, and Streamlit for the front-end.

The example receives a user’s prompt, generates a SQL query using in-memory vector database and few-shot examples. We then run the query using SQLite database and display query results in the user interface.

For simplicity, we use the in-memory Chroma database to store and search for embeddings vectors. In a real-world scenario at scale, you will likely want to use a persistent data store like the vector engine for Amazon OpenSearch Serverless or the pgvector extension for PostgreSQL.

Contents

The example consists of four files:

• Streamlit application in Python
• Supporting file to make calls to Bedrock to run the SQL chain
• Requirements file, and a data file to search against
• SQLite helper file to run queries
• Local SQLite Northwind database

Requirements

You need an AWS account with following Bedrock models enabled;

• amazon.titan-embed-text-v1
• anthropic.claude-v2

You need to setup your AWS profile and setup your "default" profile with the AWS account credentials where you have above Bedrock models enabled

Setup

From the command line, run the following in the code folder:

pip3 install -r requirements.txt -U

Running

From the command line, run the following in the code folder:

streamlit run sql_chat_ui.py

You should now be able to access the Streamlit web application from your browser.

Try a few prompts from the web application:

• Can you list customers which placed orders in the last 30 days?
• Can you get alphabetical list of products?
• For each order, calculate a subtotal for each Order (identified by OrderID)?
• For each employee, can you get their sales amount, broken down by country name?
• Can you calculate sales price for each order after discount is applied?
• For each category, can you get the list of products sold and the total sales amount?
• Can you list ten most expensive products?
• Can you list products above average price?
• Can you show sales amount for each quarter?
• Can you list number of units in stock by category and supplier continent?
• What are the total sales amounts by year?
• What are the top 5 most expensive products?
• What customers have spent over $1000 in total?
• What products were sold in the last month?
• What is the total revenue for each employee?