An AI engine is the essential component that drives an AI agent. It handles user input—text and voice, understands the user intent, and generates the appropriate responses.

Administrators can choose the AI engine that best aligns with how they want their scripted AI agents to behave.

Components of an AI engine

The core components are relevant for scripted AI agents, but their function is slightly different:

  • Natural Language Understanding (NLU): The NLU component maps customer input to the intents defined in their script. It recognizes the customer's input within the framework.
  • Dialogue Management: This component manages the flow of the conversation according to the script. It ensures the agent follows the defined paths and provides the correct responses based on the recognized intent and context.
  • Response Retrieval: This component delivers the responses configured in the script.

For voice-based interactions, in addition to the components mentioned above, the AI engine also includes ASR (Automatic Speech Recognition) and TTS (Text-to-Speech).

How to choose the right engine

  • Webex AI Pro 1.0 (with Swiftmatch)

    This AI engine is ideal for developing AI agents that manage diverse user expressions while accurately mapping inputs to predefined intents, ensuring consistent and reliable interactions across various scenarios. It's useful for:

    • Handling smaller training data set: If the training data set has fewer than 10 utterances per intent, this engine is more suitable.
    • Multilingual scripted agents: This is a good choice for creating agents that handle conversations in multiple languages.
    • Scripted agents with some level of "smart matching": While the responses are scripted, this AI engine offers a natural feel by matching user input to the closest intent, even if the phrasing isn't exact.

    Benefits: Helps with input variations, better with smaller training data sets, has multilingual support, supports smart matching.

    Limitations: Swiftmatch excels in its strong natural language understanding abilities. However, if your script requires flexibility in matching user input to intents—allowing for variations in phrasing—Swiftmatch might need extra data configuration with diverse training data. It's designed for precise and rigid matching, which can make handling variations more challenging.

  • Webex AI Pro 1.0 (with MindMeld)

    This AI engine is well suited for scripted agents with complex, multistep conversational flows. It's useful for:

    • Handling medium training data set: If the training data set has fewer than 20 utterances per intent, this engine is more suitable.
    • Complex scripted workflows: If the script involves multiple steps, conditions, or branching logic, MindMeld can manage those complex flows.
    • Multilingual scripted agents: MindMeld supports multiple languages.
    • Performs role and entity classification: Enhances the understanding of user input by identifying sentence parts and important details, improving the accuracy and relevance of responses.

    Benefits: Good for complex scripted flows, good for medium data sets, and has multilingual support, more accurate than RASA, straightforward and efficient choice for basic intent classification.

    Limitations: MindMeld can handle complex flows within the platform's existing limitations, but it’s suitable towards structured conversations. If a script demands highly flexible and dynamic dialogue management, with conversation flows that can significantly shift based on customer input, MindMeld's dependence on predefined flows might be limiting. It’s less suitable for free-form and unpredictable conversations, even in a scripted setting.

  • Webex AI Pro 1.0 (with RASA)

    This AI engine is well suited for building scripted agents that require flexible, multistep conversational flows with its customizable dialogue management and robust AI capabilities. It's useful for:

    • Handling large training data set: If the training data set has more than 25 utterances per intent, this engine is more suitable.
    • Performs role and entity classification: Enhances the understanding of customer input by identifying sentence parts and important details, improving the accuracy and relevance of responses.

    Benefits: Displays accurate results, suitable for simple to complex scripts, good for large training data sets, and more complex scenarios.

    Limitations: Needs more training data; takes much time to train.

AI engine brings together speech technology (ASR/TTS), Large Language Models (LLMs), intelligent guardrails, and expertly crafted system prompts together into one single selection choice on AI Agent Studio.

When creating a new AI agent, you can choose from multiple AI engines tailored to meet their unique needs.

Autonomous AI agent currently offers two AI engine selection options:

  1. Webex AI Pro 1.0: Ideal for most contact center use cases with global language support and human-like interactions. To view the list of supported languages and voices, see the Supported languages and voices article.
  2. Webex AI Pro-US 1.0: Perfect for scenarios requiring an enhanced human-like conversational experience, available in English only.

How to choose the right engine

  • Language support : Webex AI Pro-US 1.0 is available for English only while Webex AI Pro 1.0 supports English + various other languages in Beta.
  • Geo restrictions: Webex AI Pro-US 1.0 is available only for US customers while Webex AI Pro 1.0 is globally available.
  • Voice Experience: Webex AI Pro-US 1.0 offers an enhanced human-like conversational experience but is limited to fewer voices while Webex AI Pro 1.0 offers a wide range of voices for human-like interactions in various accents.
  • Interim messages: Webex AI Pro-US 1.0 is optimized to generate quicker interim messages using proprietary models. Webex AI Pro 1.0 generates more contextual interim responses using LLMs.