Gen AI ChatBot Tutorial - Amazon Connect, Amazon Lex and Knowledge bases for Amazon Bedrock

Architecture and Tutorial on building a completely Automated Customer Support using Generative AI. We use Amazon Connect, Lex, Bedrock Knowledge Base to build this solution.Data Source is a product catalog pdf file in a S3 bucket, and the AI Customer Support answers questions based on that. You can build a Gen AI Call Center using this no-code solution.
Timelines:
00:23 Architecture
00:31 Tutorial Begins
00:38 Data Source
00:50 Products Catalog
01:00 Create Knowledge Base
02:26 Create Lex Bot
04:18 Create Gen AI Intent
07:48 Create Connect Instance
09:27 Login to Connect Instance
09:52 Create Flow
12:01 Add Phone Number to Connect Instance
12:40 Add Communication Widget to Connect Instance
13:44 Add Widget to HTML page
14:02 Final Demo!
#aichatbot #generativeai #aws #bedrock #genai #chatbot #lex #amazonlex #knowledgebase #amazonq

[@harishkumars3209]

Very nice work..

[@sairashid]

Great video. We’re building a similar solution and would like your insights in a consulting capacity. How can I reach you?

[@juanpa9590]

Amazing tutorial, thank you!!
Maybe a dumb question but can I asign a phone number to the flow that you made for example and have and interaction with the bot using that IA intent playing in prompt the answer of the model bot?

[@ArchitectureBytes]

In the tutorial - Phone number is attached to the Amazon Connect Flow (ProductFlow) - which in turn uses Lex Bot.
I guess you are asking - if we can we assign Phone Number directly to the Lex Bot (or an Intent)? As of today - I don't see a way to attach a Phone Number directly to a Lex Bot.

[@juanpa9590]

@@ArchitectureBytesWell actually what I wanted to see is if there is some way to ask about the products just like you did in the example but through a call. Carching the customer inputs and reproducing the bot responses until you get the goodbyeIntent but all of this in a phone call

[@ArchitectureBytes]

@@juanpa9590 Yes of course - pls see the demo at 14:55 - Customer Support via Phone Call!

[@chrismiller3591]

In the QnA Intent, you didn't have a next step? I didn't see one in closing response and the fulfillment box was closed. Thanks for clarifying.

[@ArchitectureBytes]

The QnA Intent (ProductIntent) is a built-in Gen AI intent, it can answer questions automatically from the Knowledge Base using a foundation model (LLM)

[@chrismiller3591]

@@ArchitectureBytes @ 5:17 in the video, you have the option to click into the fulfillment area and designate what you want to happen as a next step in the Success Response and others. For example, you can Wait for Users Input or go to Intent, or go to Slot within and Intent, to name a few options. I'm curious about what you've selected because I have the same intent created. Is your next step truly set on nothing? If it's set on Wait for Users Input, I found it interesting that when the user said Thanks, it clicked over to the good bye intent. That's all great and designed well. I'm just curious how you get the QnA intent to break out of AI mode and get back to pre-made intents. Thanks

[@ArchitectureBytes]

Ok I understand your question now.
For the ProductIntent (QnA Intent) I have not set any specific options in Fulfillment or Closing Response, and there is no Initial Response or Slot options available here.
My understanding is that, each user prompt is evaluated independently every-time to determine the most appropriate intent to handle it.
However there are scenarios where conversation can get "stuck" - particularly if the intent's fulfillment requires multiple steps or slots to be filled. In such cases, the chatbot may expect the user to provide specific information to complete the current intent before moving on to another one.
And if the default behavior is not the best, you can guide the conversation in desired direction by providing 'Next step in Conversation'