A first look at Vertex AI Conversation (aka Gen App Builder Conversational AI) - Live demo
ฝัง
- เผยแพร่เมื่อ 1 ก.ย. 2023
- This video walks through the process of creating a generative AI powered chatbot with Vertex AI Conversation on Google Cloud Platform.
In the demo, I show how to use a website as a data store on which to ground the bot's responses. Then, I show how to customize the LLM prompt to steer the model's responses.
Product announcement: cloud.google.com/blog/product...
Documentation: cloud.google.com/generative-a...
WOW Victor this is amazing Thanks for the demo :)
thanks for the demo Victor. the agent ingests/scrapes data from multiple pages. as an end consumer of the chatbot, when i get a chat response, is it possible to find out what page the response is from? meaning that if i'd like to learn more, can i go to the page where the citation comes from? thanks
Yes, the way it works is that a URL to the source page (or document, in case you'd use unstructured data as the source) is provided in the response. In addition, Google will even highlight the parts of the text in the page when you open to show where the answers are coming from. Might only work with the chrome browser, not sure.
Hope this helps!
Hi Victor.. is it possible to add a video on how we can use bigquery datasets to create such search and conversation app...
How did you get public web content crawl?
I didn't find it anywhere 😕
I used my own example question to train a bison model in the generative ai studio. Is it possible to both use Dialogflow and use your own questions to train a chatbot? Great video, thanks for the explainer :)
If by own questions you mean a prompt you can send to the model, yes :)
couple of questions:
1. when you create a website datastore, did it not prompt you to verify the domain? it needed to verify for me.
2. I see youa re asetting a prompt for generative fallback, what about the Data store prompt does that matter?
1. Yes, domain verification is required. I did not show this in the demo.
2. There's no way to set any system prompt, really only the generative fallback (and you can also set it for "generators" which is the same concept as generative fallback but it applies to specific intents, i.e., how to have the model answer or behave for specific questions or utterances matching an intent)
Hi Victor… how did you embed the dialog flow agent to the site? That’s the key…
If you click on Publish you'll have a javascript widget you can copy and paste into any website
thanks for the video Vctor, this is amazing, Just wondering how did you create the website there please at the end?
basically just a click of a button. that generates the embedding code you can place on any website you can edit.
Thanks so lot :), Is it possible to adjust the chat to appear bit bigger in rectangular shape, I'm basically planning to create a website just for this, so users can get information about a data, is this the best? or is there a alternative ways. @@chrsl3
Hello, I have a question. If I am working with a team, is it one person who is supposed to have a veryExAI account that will be charged? Or will each user, though sharing the same project, be charged in their account?
A Vertex AI Conversation app is charged per query, so it doesn't matter how many users there are, only how much usage there is of the app itself. It's a single bill that goes to the billing account linked to the project. If multiple projects are linked to the same billing account (which is common), all billing is consolidated there and paid by whatever credit card is associated with that billing account. Hope that helps!
Hi Victor, how can this bot email the conversation to a recepient at the end. Thanks.
Yes it can! Just need to use a mail API
Thanks Victor, this look cool, can you make a more detail video, step by step how you training this agent?
The agent is trained under the hood when you create an app in Vertex AI Conversation, so there's nothing extra I've done. More videos on Vertex AI Conversation to come soon though :)
👍