Inaccurate results from Conversation AI

[sakshi_gor]

Hello, I'm building a chatbot using Vertex AI conversation and Dialogflow CX. for the chatbox the only datastore is a domainset (URLs). Almost everything else is working fine and it is giving relevant information about almost everything, but when it comes to pricing, it does not give even relevant information. I am using the default generative model and Advanced NLU with 0.75 threshold. Earlier it was an image, and since I always want my chatbot to fetch data from my website, I changed it to simple short text, for example item (currency unit) - value (unit). I also added the data in a table, but nothing is working for me. Also I have a question, Can our model perform OCR (Optical character recognition) too?

Hello @Poala_Tenorio if you have any idea what I can do better to make it right.

[xavidop]

Hi!

Is the data store fully indexed? make sure that it is fully indexed.

Make sure that the data store is being triggered, here you have the order of evaluation:

And finally, OCR is not an option for data store. You have to upload websites, documents or structured/unstructured text data

Best,

Xavi

[sakshi_gor]

Hello Xavi!

Thank you for responding.

Yes, my links are fully indexed, also I am getting the response from the same page the pricing is mentioned. Still it does not give the rate but says something which is not relevant.

Also for OCR, I meant if there is any image on a webpage, can it perform OCR and understand that too?

Thank you,

Sakshi

[xavidop]

That is weird, It should be working. I do not know why is it not working fine. Sorry for not helping you on that front.

I know that Vertex AI Search & Conversation does not perform OCR, so you will need to perform that on your own and upload the info to the datastore.

Xavi

[sakshi_gor]

Thank you for your response! I really appreciate it. 

Best,

Sakshi

[nabelou10]

I get really bad answers too, I have tried Vertex AI search and search with grounding, it is painfully inaccurate.