This is a simple tool to search for data in CSV files, aggregating results into a single result file with references back to the original files. It is designed to be executed from a DevContainer and utilize an Azure OpenAI deployed model to take a user's prompt and search record by record to find the best matches.
The minimum requirements to run this tool are:
- This repository cloned to your local machine
- An Azure OpenAI endpoint and deployed model (and access to the endpoint and API Key) - see below on using Bicep to deploy an instance
- VS Code and Docker to run the DevContainer for this solution
Copy the .envsample to .env and update with your Azure values; all values
are required to complete the deployment and to run the application.
| Variable | Description | Recommended Value |
|---|---|---|
AZURE_LOCATION |
The region the Azure resources will deploy into | eastus |
AZURE_APPENV |
A unique "workload" name; Azure resources will have this as part of their names | |
AZURE_SUBSCRIPTIONID |
The GUID of your Azure subscription | |
AZURE_OPENAI_MODELNAME |
The model to use in your Azure OpenAI deployment | gpt-35-turbo-16k |
AZURE_OPENAI_MODELVERSION |
The version of the model to use in your Azure OpenAI deployment | 0613 |
From the DevContainer VS Code terminal, run the ./deploy.sh script to
automatically deploy a new resource group and Azure OpenAI instance to your
Azure subscription. The process will take a few minutes. If there are any errors
during the deployments, logs will be written to the ./logs directory and
timestamped based on the run.
Once the deployment has completed successfully, you can now run the application.
If you are done with the resources and this application, running the
./remove.sh script will remove all Azure resources created by the deployment.
You can either run the application in the VS Code debugger by launching the
Python Debugger: Streamlit App task, or by utilizing the ./run.sh script in
the root of the repository. Utilizing the debugger will allow you to breakpoint
and see the steps of the application, while running the script will be bound to
just the web app.
Once you have started the application, you can view it in the browser at http://localhost:8501.
If you have issues utilizing Azure OpenAI, check that you are logged in to the
az cli by opening a terminal in VS Code and running az login. This will
ensure that the application can authenticate to Azure on your behalf.
CSV files should be placed in the _data/_input directory or uploaded through
the web app; all files that are in this directory will be searched as part of
the run.
There are only a few requirements for the CSV files:
- They must be valid (e.g., escaped values if necessary)
- There must be a header row
- The header row much have two columns,
IdandName - Any other columns will be included as part of the search
Output will be visible in the web application. Use the Export to Disk button
to save the output to disk.The output file be placed in the _data/_output
directory. You can also download the output file directly from the data grid on
the web app without saving it to the repository directory.