How to use Gemini pro 1.5 for dialogflow cx conversation summarization

[Vinoth1097]

Hi Team,

I have a CX agent which asks series of questions(like 10) to the user which are of yes/no type. At the end of the conversation, we are trying to show a conversation summary to the user. Currently im utilizing Generators to create a summarization with below prompt and im using text-bison@latest model with (temperature: 0.2, token limit: 1024, top-k: 40, top-p: 0.95)

Prompt:
You are an expert at summarizing conversations between a User($username) and an Agent. When summarizing, always use the "$username" instead of "User" or "the user".

Always format your summary as follows: "**Summary**: "
Summarize below conversation in 8 to 10 sentences(ensure its not less than 1000 words). Do not add extra words.
Conversation {$conversation, user: $last-user-utterance}
Summary:

Requirement:
I have tried with several available models in generators which are not giving best results. I was trying out Gemini 1.5 preview and it gives better results. Is there an option to send the conversation history from dialogflow in a api call and use gemini 1.5 for summarization? If that's not feasible, can you suggest best model with configuration options along with llm prompt.

[xavidop]

Hi, you can call a webhook that makes that call to Gemini 1.5.

You can retrieve the history from the webhook by accessing bigquery. You will need to check if you can access these variables from the webhook: $conversation, $last-user-utterance. if you can do that, then you do not need to go to bigquery.

another solution could be to open a support ticket or ping your GCP customer manager and see what they can do.

Best,

Xavi