How to fix variablility in responses of the dialogflow playbook fully generative agent?

I have created a fully generative single playbook agent on dialogflow (Conversational Agents).

The responses that the agent gives to the same questions is different when I test it at two different point of times.

I am testing my agent with a very small sample dataset. It looks something like this :

A, B, C, D, E are column names

There are some empty values and some duplicate values also in the column.

There are some questions that I repeatedly test the agent with. For example,

Q1 : What is the average value under column E when the value under column D is d2.

Q2 : List all the values under column B when the date is yyyy-mm-dd

Q3 : What is the value under column C when the value under column D is d1 and the column A value is yyyy-mm-dd.

For Q1, when I tested the agent a few days ago, it was giving the correct answer. But when I tested the same agent without making any changes to it today, the agent responded by saying it does not have access to the data.

For Q2, the agent is never able to filter other columns based on the date column. However, if I ask the agent to tell me what is the date when the value under column B is b1, it is able to output the date in the response for me.

For Q3, Note that this question requires the agent to filter based on two columns and give answer from the third column.

I had enabled the following API’s for my project yesterday (from here ):

  • Vertex AI API

  • Cloud Storage API

  • Cloud Logging API

  • Cloud Monitoring API

  • Cloud Trace API

When I had enabled these api’s yesterday, I tested the agent and it was able to answer Q3. But today again the agent is unable to answer Q3. Also, there is no way to confirm if the API’s are enabled when I am testing my agent. I am able to enable the api’s again and agian everytime I visit the link where I enabled them from earlier.

I have even tried increasing the input and output token limit in attempt to fix this issue. But it did not help.

  • Input token limit : up to 32k
  • Output token limit : upto 1024

How do I fix this variablity in the agent’s responses?

@Gaurangi_Garg Can You also explain how you integrate this data Playbook
Method 1 or Method 2

I have used method 1 (the blue colored diagram)