Generative AI

The inGenious AI platform offers multiple generative AI features to improve your chatbot experience and help you maintain your chatbot content, such as improving your NLP question matching, generating training phrase suggestions, or generating handover transcript summaries for your livechat agents. These features use Large Language Models (LLMs) to generate the results.

You have full control over which generative AI features you enable and use, which model you use for each feature, and the prompt that is used to generate the result.

Generative AI features must be enabled for your chatbot by inGenious AI support before you can use them, and can only be configured by users with an administrator or publisher role in your team. See Generative AI permissions.

In the inGenious AI platform, generative AI features are powered by:

• LLM Connectors, which connect your chatbot to generative AI models.
  Different models may have different costs and support different tasks. The inGenious AI platform supports multiple generative AI providers and models, and you can connect your chatbot to as many as you need.
• LLM Usages, which define how LLMs function with the inGenious AI platform and your live chatbot.
  Each usage type serves a specific purpose and contains one or more smaller tasks. You can create multiple usages of the same type, but the chatbot only uses the primary usage. 
• LLM Tasks, which define the prompts, connector, and configurations to use for a specific generative AI feature. You can customise the prompts to suit your needs.

You can also create guardrails to help you monitor and manage undesired content, annotations to assist the model, and use automated testing to improve the prompts of some tasks.

Guardrails

Guardrails grant you more control over the content that is sent to and generated by LLMs. Use guardrails to check for harmful topics, profane or undesirable words, and sensitive information like email addresses, as well as to check how relevant and accurate a generated response is.

When a guardrail filter is triggered, you can flag the content but allow the chatbot to continue as normal, if you want to monitor certain content without impacting the chatbot experience, or you can change how the chatbot responds, such as starting a passage for the type of content detected.

Annotations

LLM models often achieve better results when they have more specific information to work with. You can use annotations to add context to conversations and passages that wouldn't be available from the chat history or disambiguate passages with similar content. For example:

• You could add annotations to more thoroughly describe the issue a passage solves or explain some terms that your organisation uses in an unusual way. 
• If you have several passages with very similar content that apply to different situations, such as a 'close card' passage for different card types, you can add annotations to identify the card type without changing the passage content. 

The annotation is never visible to your chatbot users.

Annotations are only used by some tasks, and only if you configure the tasks to use them. 

Automated tests

Generative AI permissions