Copilots

Karini AI's Copilots are the innovative end-user applications or chat interfaces that utilize the pre-built recipes as the underlying engines for conversations with end users. You can design copilots that enforce your organizations communications guideline, represent your brand and record essential feedbacks from the users.

For details about designing the copilot interface, refer to Export Recipe.

Key capabilities of Karini AI's copilots:

Seamless access:

Users can access Copilots from within Karini AI's platform for which they must have Copilots user, Administrator or Superuser role assigned to them. You can also incorporate the copilot within your own domain url and allow your domain users access the copilot interface.

Engaging in Dialogue:

Copilot enables users to pose questions and receive responses, fostering interactive engagement. Karini AI's copilots also supports the streaming experience for almost all SOTA model providers, citations (references), and optional follow-up questions. The feature is vital as it delivers a uniform user experience when enterprises switch between model providers for A/B testing and model upgrades.

Feedback Mechanisms:

Karini AI's Copilot enables you to capture comprehensive feedback from copilot users along with the basic upvote, downvote. You can capture following types of custom feedbacks.

• 5-stars: Copilot users can rate the content with a 5-star rating.

• option: You can configure options to capture feedback on Copilot interactions based your specific requirement. E.g. You may configure the options as "Inaccurate", "Vaguely accurate", "Accurate". These predefined options are made available to the copilot users to provide feedback on Copilot interactions.

• text feedback: Copilot users can provide descriptive text feedback.

• Correction: Enabling this option will allow the copilot users to provide corrections or improvements to the copilot response.

Detecting Greeting or Non-Specific questions:

Karini AI's copilot gives you the flexibility to detect irrelevant or un-related user questions and respond with an appropriate message to engage with users in a positive way and encourage them to ask questions related to the topic.

Content Safety Filters:

You can enable content safety filters in the recipe to prevent harmful or toxic content from being accepted or generated in the copilot. You have the flexibilty to respond to such content according to your organizations content safety messaging policy.

Copy Functionality:

Users can easily copy both questions and answers, facilitating reference and information sharing.

Copilot History:

You can access the entire chat history of a copilot which is useful for auditing, cost and usage monitoring and performance improvement. You can also derive trends and usage patterns based on the copilot history.

In essence, Copilot represents a powerful platform for interactive engagement, enriched by features such as suggested follow-ups questions, feedback mechanisms, and correction capabilities, ultimately fostering continuous learning and refinement.

<<Copilot Video>>>

Copilot Actions

On the copilots list page, you can utilize various the Actions button to carry out the following set of actions. Only Administrator and Superuser roles can view and perform these actions.

Edit Recipe

This option redirects you to the underlying Recipe of the Copilot, where you can make necessary adjustments to the recipe pipeline. You must publish the recipe in order for any recipe changes to be reflected in the copilot. You can also create multiple exports (and hence multiple Copilots) from the same recipe as your use case demands.

Edit Export

This option redirects you to the Export section associated with the Copilot, where you can make necessary edits to the formatting, layout, or any other aspects related to export. You must save the export form in order to make the changes reflected in the copilot for the new sessions.

View History

This option redirects you to the Copilot History page which includes comprehensive Dashboard about the copilot activity and a table containing detailed information about the copilot user interactions.

• User name: User name of the person interacting with the copilot.

• API Session: Session id of the copilot.

• Recipe name: Name of the recipe associated with the copilot.

• Recipe version: Version of the recipe utilized by the copilot.

• Chatbot name: Copilot name.

• Status: Status of the responses (e.g. success, aborted, unsafe, greeting).

• Question: Query posed by user.

• Answer: Response provided by the copilot.

• Response Type: Type of response (e.g. copilot)

• Prompt: Name of the prompt used in recipe.

• Feedback: Feedback provided by the user.

• Tokens: Number of tokens utilized in generating the response.

View Traces

For each record in the copilot history table, you can view traces to get insights into the detailed step-by-step information about the request processing and response generation. Trace has two sections as Prompt and Attributes.

1. Prompt: You can view the traces of each operation executed during the processing . It includes the following:

   • Input

   • Output

2. Attributes: These include various parameters and metrics associated with each request. Some of the attributes include:

   • Input Tokens

   • Completion tokens

   • Model parameters such as temperature, max tokens etc.

   

Export Dataset

Copilot history hold valuable information about the user interactions with your copilot, such as the sessions, users, questions, answers, feedback, prompt etc. You can select the record and export this information to use in further downstream processing or auditing, or even to fine-tune the model if required.