4.6. Agent Training

General training screen

To proceed to agent training, you need to open the section of the same name within the platform's control panel:

This section contains the following information:

• Analytical summary regarding training needs.
• Information on the agent’s recent training tasks, including performance evaluations.
• Section for training examples.
• Section for datasets.
• Section for proposed improvements.
• Version history of system instructions, with the option to roll back.

Starting training

Now, let’s take a look at how the agent training process works. To begin, go to the examples section:

In this section, you can add or modify various message templates, as well as approve automated correction suggestions provided by human operators or AI agents. You can also import recent problematic conversations here to determine how they should be handled or addressed.

When adding an example, you need to specify the user input (question) and the expected response from the agent, as well as select the example type and provide a moderator comment. You can then approve the example, moving it either to the drafts or to the set of active examples:

If necessary, you can add additional training sets, which can be either global or local (specific to a single agent):

In the section for suggested improvements, you can review and approve changes by comparing system instructions and the effectiveness of responses to questions against the new data:

And in case you find, after some time, that you need to return to previous system instructions, you can always roll back to a specific version:

Thus, with the help of this new tool, you can further enhance the accuracy of your agents’ performance for any purpose.