Datastore in Vertex Ai Agents

[Bhaven]

Does Datastore provide any confidence or generator score for the responses generated in Dialogflow CX?
I need to classify the responses based on their accuracy confidence levels. For example:

• Very High: Datastore is highly confident that the response is accurate.
• Good: Datastore is somewhat confident in the accuracy of the response.
• Very Low: Datastore is confident that the response is not accurate.

[pedropcamellon]

Hi! Using the web platform or API? 

[Bhaven]

In the Web Platform , I guess we can see the inside the original response for score -

"generativePrediction": {
"groundingInfo": {
"accepted": true,
"score": "VERY_HIGH",
"snippetsSupportingIndices": [
{
"indices": [
1,
2,
0
]},]}}

Wondering do we get same/similiar kind of score of response via API  in response body?

[pedropcamellon]

You get a lot of info from the API. I was trying but with my current setup I'm not seeing the snippets section. I will let you know if I get them.

Try this:

```python

def detect_intent_texts(agent, session_id, text, language_code="en-us"😞

    """

    Returns the result of detect intent with texts as inputs.

    Using the same `session_id` between requests allows continuation

    of the conversation.

    """

    session_path = f"{agent}/sessions/{session_id}"

    client_options = None

    agent_components = AgentsClient.parse_agent_path(agent)

    location_id = agent_components["location"]

    if location_id != "global":

        api_endpoint = f"{location_id}-dialogflow.googleapis.com:443"

        print(f"API Endpoint: {api_endpoint}\n")

        client_options = {"api_endpoint": api_endpoint}

    session_client = SessionsClient(client_options=client_options)

    # Prepare request

    text_input = session.TextInput(text=text)

    query_input = session.QueryInput(text=text_input, language_code=language_code)

    request = session.DetectIntentRequest(

        session=session_path, query_input=query_input

    )

    intent_detection_response = session_client.detect_intent(request=request)

    response_json = MessageToDict(intent_detection_response._pb)

    response_messages = response_json["queryResult"]["responseMessages"]

    response_text = response_messages[0]["text"]["text"][0]

    return response_text

```

[Bhaven · Accepted solution]

To be able to see the Snippets Section in Web Platform, I believe you need to add datastore handler in Start or Any Preferred Page in Dialogflow CX.
Try this - Datastore Agents in Dialogflow CX 
* * To be able to use datastore inside dialogflow CX, you need to enable Vertex AI API in console.**

Did some R&D and got to know the searchSnippets is not logged inside response body when calling dialogflow via API.