Admin: Basic selection rule AI integration

RPI can use Artificial Intelligence (AI) to assist in the building of criteria within basic selection rules.

RPI’s usage of AI relies on Azure OpenAI and Azure Search services:

• https://azure.microsoft.com/en-us/products/ai-services/openai-service

• https://azure.microsoft.com/en-us/products/ai-services/ai-search

The Azure OpenAI models used by RPI are:

• OpenAI chat models GPT-35-Turbo or GPT-4

• text-embedding-ada-002

The following are also required:

• Azure Blob Storage account

• Here is a sample of the default vector algorithm configuration which is suitable for the purposes of AI rule building in RPI:

Use of Azure services in this context is subject to cost, and is determined by the selected model type as well as the complexity of a given prompt and the resulting returned attributes. Please refer to the following for more information on pricing: https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/

Azure OpenAI is a proprietary service of Microsoft Corporation and requires a separate license subscription directly from Microsoft Corporation to access.

The following application settings must be configured on both the execution service and Interaction API containers to use this feature:

• RPI__NLP__ApiKey=[Azure OpenAI APIKey]

• RPI__NLP__ApiBase= https://[Azure OpenAI Service Name].openai.azure.com/

• RPI__NLP__ApiVersion=[Azure OpenAI API Version]

• RPI__NLP__ChatGptEngine[Azure OpenAI API Chat Model Deployment]

• RPI__NLP__ChatGptTemp=[Azure OpenAI Chat Model Temperature]

• RPI__NLP__SearchEndpoint=https://[AI Search Service Name].search.windows.net

• RPI__NLP__SearchKey=[Azure Search AI Key]

• RPI__NLP__EmbeddingsModel=[Azure Embeddings Model Deployment]

• RPI__NLP__Model__ModelDimensions=1536

• RPI__NLP__Model__VectorSearchProfile=[Azure AI Search Vector Search Profile]

• RPI__NLP__Model__VectorSearchConfig=[Azure AI Search Vector Search Config]

• RPI__NLP__Model__ConnectionString=[Azure Storage Connection String]

• RPI__NLP__Model__ContainerName=[Azure Storage Container Name]

• RPI__NLP__Model__BlobFolder=[Azure Storage Blob]

Optionally, the following setting can be used to enable trace logging:

• RPI__NLP__EnableTrace [Boolean]

When set to true, at invocation of SQL Database Definition > Update AI Model, each time a call is made to the OpenAI API, the following are logged to the RPI Server Log as errors:

• JSON representation of all RPI_NLP_ settings, with keys partially obfuscated

• Endpoint called

• Payload sent

• Response received