Datastore agents answer customization in Dialogflow CX

[anonymous]

Hi,

I’m trying to implement dialogflow cx as a substitute to langchain but i cant seem to get the same performance. Seems like it doesnt extract all the information from the documents.: sometimes it gives really short answers, avoiding important parts of the information. I see there is a partially customizable prompt in the GenerativeAI options of the agent but it seem to help much.

• Aside from the boosting and filtering part i can see in the dialogflow cx docs is there a way to customize more the generative side of the agent which interprets the retrieved documents (in the way i would do with a prompt) ?
• Should i expect a worse performance w.r.t. langchain in the case of a less clear/organized datasource?

Thank you!

[xavidop]

Hi Consta!

The configuration you said is the only configuration you can do at Dialogflow CX:

It uses Vertex AI Search and Conversation under the hood, you should not expect worse performance. If you want to use another technology but still using Dialogflow CX as a conversational AI manager I would suggest calling a webhook that retrieves that info from an external system.

Best,

Xavi