Dialogflow CX data store FAQ

[rob127]

Hi , using datasore .csv for FAQ Documants and using no other data source I can see on one document id the following

{ "Title": "flyrod start", "Answer": "For beginners affordable is priority", "Question": "I'm a beginner. What's a good fly rod to start with?", "Link": "https://www.flyfishingnewbies.com/best-fly-rods-for-beginners/" }

so I try this in chat agent using qestion ''m a beginner. What's a good fly rod to start with?' and get back a $request.generative.response and not the answer in my csv. Is there some setting to make sure the data store is used first? 

[rob127]

Has anyone actually got the FAQ documents data source to actually work or is this still in a beta stage ?

[thesammiller]

Yes, I was able to get this to work. Columns 'question' and 'answer' are case sensitive to lowercase. Import into Big Query, then into a Data Store. 

[CosminB]

Does the faq structured document in csv format also works without uploading it first to big query? I simply cannot get it to provide me the links inside the csv 

[codie_petersen]

The response is an object, you will have to pull the relevant data out of the object. I'm not sure what language you are using but here is an implementation for my project:

def _parse_response(self, response: GenerationResponse) -> LlmResponse:

        """

        Parse the response from the Vertex AI API and return an LlmResponse
        object.

        """

        llm_response = LlmResponse()

        llm_response.id = self._response_id

        llm_response.type = "vertexai-generative-model"

        llm_response.model = self.settings.model

        llm_response.created = int(time.time())

        choice = Choice()

        #Safety check

        response_details = response.candidates.copy()

        raise_exception = False

        blocked_reasons = []

        for detail in response_details:

            if detail.finish_reason == FinishReason.SAFETY:

                for rating in detail.safety_ratings:

                    if rating.blocked:

                        raise_exception = True

                        blocked_reasons.append(str(rating.category))

        if raise_exception:

            print(response_details)

            message = f"Blocked by safety filter.\nReasons:"

            for i, reason in enumerate(blocked_reasons😞

                message += f"\n\t{i}). {reason}"

            raise Exception(message)

        choice.content = response.text

        llm_response.k_choices = [choice]

        return llm_response

I've highlighted how I deal with the response object in regard to the actual text retrieval. The text is what you want to print, not the response object.