Dialogflow CX Agent Not Using Generative AI (Gen App Builder) for Product Search

[harishhrk]

Hello everyone,

I'm working on building a shopping assistant chatbot using Dialogflow CX and integrating it with a Gen App Builder Chat App (Vertex AI Search and Conversation Data Store) to handle product search queries. I'm following the standard process but am stuck on getting the generative AI to actually provide responses.

My Goal: To have my Dialogflow CX agent understand product-related queries (e.g., "I want to buy a toner," "Do you have any shampoo?") and then use my connected Gen App Builder data store to search for products and generate a natural language response based on the product data.

What I've Set Up So Far:

1. Gen App Builder Chat App (Data Store): I have a Gen App Builder app named glamzy-products which is connected to my product data. I've successfully tested this app independently within Gen App Builder, and it can search my data.

2. Dialogflow CX Agent:

   • Created a Dialogflow CX agent (glamzy-shopping-agent).

   • Created a find_product intent with various training phrases (e.g., "I'm looking for a shampoo," "Do you have any Chik products?").

   • On the Default Start Flow -> Start Page:

     • Under the "Data stores" section, I have successfully linked and see my glamzy-products data store listed.

     • Under the "Generators" section, I have created and configured a generator named glamzy-product-recommender with a detailed prompt that guides the AI on how to respond and format product information (e.g., "You are a friendly and helpful shopping assistant... search the connected data store and recommend up to 3 relevant products...").

     • Crucially, I have ensured there are NO static text responses in the "Agent responses" area of the Start Page fulfillment.

     • I have also confirmed that there are NO static text responses in the fulfillment section of my find_product intent. The intent's fulfillment is intentionally empty so it should fall back to the generative response.

The Problem:

When I test the agent in the "Test Agent" panel and type queries like "Shampoo" or "I want to buy a toner," the agent consistently responds with:

Agent Says: I'm sorry, I didn't get that. Can you rephrase your question?

This indicates that:

• The find_product intent might not be matching consistently, OR

• Even if the intent matches (or if it's a no-match scenario), the generative AI from my glamzy-products data store is not being triggered or used for the response.

I'm effectively getting a default "no match" response, despite having the generator and data store configured.

My Questions:

1. Given my setup, what could be preventing the Dialogflow CX agent from using the glamzy-product-recommender generator to provide a response from the glamzy-products data store?

2. I understand there are "Generators" sections on both the Start Page and potentially within specific intent routes. Where exactly should I configure the use of my glamzy-product-recommender (i.e., by adding a "Generative" response type) to ensure it acts as the primary fulfillment for product queries? Should it be in the Start Page's fulfillment or directly within the find_product intent's route fulfillment?

3. Are there any specific "enable generative fallback" checkboxes or settings I might be missing in the latest Dialogflow CX UI that would prevent the generator from being used, even if linked and defined? (I've looked for an explicit one under the Data Stores section but it's not present there directly anymore).

Any guidance or troubleshooting tips would be greatly appreciated! Thank you.
1. Below is my data format for 1 single product i ahve added almost 12k products in the cloud.

{"id": "0008ff72-dacd-41ff-b85e-45b9323e198e", "structData": {"name": "Chik Shampoo, Black, 80ml", "description": "Chik Shampoo, Black, 80ml - Chik product", "category": "Shampoo", "brand": "Chik", "product_id": "cb9509bf-c337-4503-bff3-b594aaced1c9", "variant_id": "0008ff72-dacd-41ff-b85e-45b9323e198e", "availability": "out_of_stock", "image_url": "https://be-ui-uploads.s3.ap-south-1.amazonaws.com/production/d486f9be258ecedd3579faed415014af/chik-thick-glossy-80ml.jpg%20%281%29.jpg", "product_url": "https://glamzy.in/home/product-detail/0008ff72-dacd-41ff-b85e-45b9323e198e", "mrp": 62}}

2. My App type is chat
3. Below is playbook screenshot and others 

[nikacalupas]

Hi harishhrk,

Welcome to the Google Cloud Community!

Let’s go through your question one by one:

1. Given my setup, what could be preventing the Dialogflow CX agent from using the glamzy-product-recommender generator to provide a response from the glamzy-products data store?

• As described above, the find_product intent route currently has an empty fulfillment. You must explicitly configure this route's fulfillment to use your generator. Simply having the generator and data store in place is not sufficient; you need to instruct Dialogflow CX to use them to generate a response for a specific scenario. Dialogflow CX requires a clear directive in its fulfillment pipeline to trigger a generative response. If a "Generative" response type is not configured, it won't know to call your generator. 

2.  I understand there are "Generators" sections on both the Start Page and potentially within specific intent routes. Where exactly should I configure the use of my glamzy-product-recommender (i.e., by adding a "Generative" response type) to ensure it acts as the primary fulfillment for product queries? Should it be in the Start Page's fulfillment or directly within the find_product intent's route fulfillment?

• You should configure it directly within the fulfillment for the find_product intent route. This is the most logical and correct place. It ensures that whenever a user expresses an intent to find a product, your specialized product generator is used.

3.  Are there any specific "enable generative fallback" checkboxes or settings I might be missing in the latest Dialogflow CX UI that would prevent the generator from being used, even if linked and defined? (I've looked for an explicit one under the Data Stores section but it's not present there directly anymore).

• The concept of "generative fallback" has been integrated into the event handlers for a page, specifically for the sys.no-match-default event. However, this is for a different purpose than what you're trying to achieve. You are aiming to fulfill a matched intent with a generative response, which requires a fix in Q2. On the other hand, generative fallback is designed to handle user input that does not match any intent route. While you're focusing on fulfilling matched intents, you can (and should) also configure a generative response for no-matches to make your agent smarter and more adaptive.

Was this helpful? If so, please accept this answer as “Solution”. If you need additional assistance, reply here within 2 business days and I’ll be happy to help.