Conversation Orchestration¶

Conversation Orchestration in DialogGPT allows you to define and manage Conversation Types while configuring separate Chunk Shortlisting and Conversation Orchestration models. You can turn Intent Events and Conversation Events on or off and choose how each event is fulfilled (by overriding the default fulfillment), ensuring precise and adaptable conversation management that you can confidently rely on.
This flexible system gives developers complete control over conversation management, ensuring the accuracy, relevance, and efficiency of user interactions.

Conversation Types¶

It allows you to include the types of conversations that you want DialogGPT to handle in user interactions. The following are the supported conversation types. Dialogs and FAQs are enabled by default and can not be disabled.

Dialogs: Dialogs are the virtual assistant's building blocks, enabling it to perform tasks and interact with users.
FAQs: Pre-defined questions and answers that provide quick and informative responses to common inquiries.
Conversation Intents: Common phrases in user interactions with the virtual assistant.
Knowledge from Search AI: Deliver answers directly from enterprise data sources for users’ questions.

To manage the conversation types, click Settings, turn the toggle on/off to enable/disable the conversation type and click Save.

Model Configurations¶

The Model Configurations card lets you choose the models for Shortlisting Relevant Chunks and Conversation Management (Orchestration).

Shortlisting Relevant Chunks: The XO Platform supports MPNet, BGE-M3, LaBSE, and E5 embedding models, with more options coming soon. You can adjust the Similarity Threshold and Proximity Threshold in the advanced settings. In most cases, the default settings work fine. The recommended model is BGE-M3
Conversation Management: The supported LLMs assist in detecting user intent and creating execution plans. You can configure advanced settings such as Temperature, Max Tokens, and Conversation History Length. The default settings are sufficient for most use cases. Support for major commercial models will be introduced in the coming months. However, you can always integrate your own model through the custom LLM integration option. Conversation Management currently supports the following LLMs:
- OpenAI GPT-4o
- OpenAI GPT-4o mini
- Azure OpenAI GPT-4o
- XO GPT – DialogGPT

To configure the model and settings, click Settings. Select the model, adjust the settings if necessary, and click Save.

Fulfillment¶

There are two types of fulfillment events: intent events and conversation events. You can modify the fulfillment behavior for each event as needed.

Intent Events¶

The following is the intent event fulfillment behavior.

Intent not Identified: Define the fulfillment when no user intent is identified in the user utterance. By default, the fallback task is triggered.
Ambiguous Intents Identified: Define the flow when ambiguous intent is identified. You can configure how the system should handle such situations.
- Present Qualified Intents (Default): Display a list of intents for the user to select or respond to. The task chosen (Dialog, FAQ, or Answer) is triggered.
- Execute Dialog task: Select the dialog task to execute.
Answer Generation: This event handles user inputs that need answers generated from Search AI documents. When Knowledge from Search AI is enabled in DialogGPT, the system activates the Answer Generation event and adds the default event handler to the bot.
Multiple Intents: Define Multi-Intent fulfillment logic when the Conversation Orchestrator detects multiple intents in a user’s utterance. Users can modify the predefined Multi-Intent fulfillment dialog, gaining greater control and flexibility. By default, the platform executes the predefined dialog (Multiintent) to handle the Multi-Intent event.

To manage an intent, click the respective settings. Define the event configuration, use the toggle to enable/disable the intent type, and click save.

Conversation Events¶

The system activates Conversation Events when you include Conversation Intents in the Conversation types. These pre-configured system features handle common phrases like holding, restarting, agent transfer, and ending a conversation. As a platform user, you can enable one or more Conversation Intents for a Conversation Event. You can select the fulfillment option for every event.

Supported Conversation Events:

Interaction Intents: Handle actions like answering from context, holding, or refusing to answer. The default fulfillment method is ‘Executing a dialog task - ConversationEvent.’
Restart Conversation: Triggers the welcome flow. The default fulfillment method is 'Ask for Confirmation' and then based on the user response, 'Restart the Conversation'.
Agent Transfer: Transition the user to a human agent. The default fulfillment method is 'Ask for Confirmation' and then based on the user response 'Execute the Dialog Task - DefaultAgentTrasfer'.
End Conversation: Manage the termination of a conversation. The default fulfillment method is 'Ask for Confirmation' and then based on the user response, 'End the Conversation'.

To manage an intent, click the respective settings. Define the event configuration, use the toggle to enable/disable the intent type, and click save.