INSTALL-3.md

Part 4 - Deploy and use the Generative AI Factory Agent

Deploy a Large Language Model (LLM) in Azure Open AI

• Deploy a base model:
  • Login to Azure AI Foundry Portal
  • Select your deployment in Shared resources > Deployments > smart-factory
  • Copy the following information in Endpoint section: Target URI and Key. We will need them in the next section.

Create an environment variable file

• Rename the file .env_template to .env
• Retrieve the environment following variables you defined in Part 1 - Provision resources (Edge and Cloud) ==> file variables.yaml:

  FACTORY_AGENT_SP_APPID
  FACTORY_AGENT_SP_SECRET
  TENANT

• Select 'Real-Time Intelligence' from the Fabric homepage.
  
• Click on Workspaces > Smart Factory
• Select the database AIO (type: KQL Database)
• Retrieve the Fabric endpoint from Overview > Query URI > click Copy URI
• Modify environment variables in .env file

  AZURE_OPENAI_ENDPOINT           = < Azure AI Foundry Portal => Target URI >
  AZURE_OPENAI_API_KEY            = < Azure AI Foundry Portal => Key >
  AZURE_OPENAI_DEPLOYMENT_NAME    = "smart-factory"
  AZURE_OPENAI_MODEL_NAME         = "gpt-4o"
  AZURE_OPENAI_DEPLOYMENT_VERSION = "2024-08-06"
  
  AZURE_AD_TENANT_ID              = < variables.yaml => TENANT >
  KUSTO_CLUSTER                   = < Microsoft Fabric => Query URI >
  KUSTO_MANAGED_IDENTITY_APP_ID   = < variables.yaml => FACTORY_AGENT_SP_APPID >
  KUSTO_MANAGED_IDENTITY_SECRET   = < variables.yaml => FACTORY_AGENT_SP_SECRET >
  KUSTO_DATABASE_NAME             = "AIO"
  KUSTO_TABLE_NAME                = "aio_gold"

Start the Factory Agent Application

• Option 1 (from command line)

  • Start a terminal from the directory
  • Execute the following commands:

    pip install -r requirements.txt
    streamlit run .\frontend.py

• Option 2 (Docker)

  • Start a terminal from the directory
  • Execute the following commands:

    docker build . -t factory-agent:v1.0
    docker run -p 8501:8501 factory-agent:v1.0

• Launch a browser with the following URL to access the application:

  http://localhost:8501/
  

• You can now query the database using Natural Language

  IMPORTANT: No actual data from the database is transmitted to the Large Language Model; only the prompt and the database schema are shared. The LLM will generate the query to be executed against the database, but it won't execute the query itself.

• Some example queries are provided.