Skip to main content
Test reviews are step-by-step checks of the AI agent’s interactions with a simulated user. In each test you compare a message that replicates a real conversation against the response generated by the agent, and rate it. All feedback is sent to the Mindo development team to improve the agent’s behavior.

Demo video

Your comments on each test are the main source of improvement for the agent. The more specific you are when describing a problem, the better the development team can fine-tune it.

Steps to complete a test review

1

Review the split screen

In the interface you’ll see, on one side, the simulated user message (which replicates a real conversation) and, on the other, the response generated by the agent. Read both to decide whether the response is appropriate.
2

Rate each interaction with like or dislike

For every response from the agent, indicate whether you like it (like) or not (dislike). You must rate every interaction in the test.
3

Describe the problem when you dislike

If you press dislike, clearly describe what the problem is (for example: “I don’t like the greeting it used”) and click send. That comment reaches the development team along with the flagged interaction.
4

Approve or disapprove the whole conversation

Once you’ve reviewed all the interactions in the test, decide whether to approve or disapprove the conversation as a whole. If you disapprove it, you can also leave a general comment explaining why.
5

Move to the next test

When you approve or disapprove a conversation, you automatically advance to the next test (for example, from test 1 to test 2).
6

Complete every test

Each review has a set number of tests. It is mandatory to complete every test to the end so the team receives the full review. If you leave the review unfinished, the feedback is not sent.