AWS AI Customer Support Chatbots – Complete Hands-On

Learn how Amazon Lex, a fully managed AWS service, builds chatbots and voice assistants with natural language understanding and speech-to-text, scaling for customer support and booking tasks.

Learn two key concepts in Amazon Lex: intents and slots. Intents define the user goal, while slots capture needed information and enable follow-up questions.

Create your first Amazon Lex order bot by defining the PlaceOrder intent, adding utterances and slots for item id and quantity, and configuring confirmation with Lambda fulfillment (ready for DynamoDB).

Discover how AWS Lambda works with Lex by running code serverlessly to validate orders, save them in a database, and handle errors, enabling behind-the-scenes order processing.

Connect the AWS CLI to your account by configuring IAM credentials, creating a user with Administrator Access, generating access keys, and running aws configure with region and JSON output.

learn how infrastructure as code uses serverless framework with node.js to define AWS resources in serverless.yaml, generate CloudFormation templates, and deploy lambda functions and APIs automatically from your local setup.

Install the serverless framework for infrastructure as code by first downloading and installing Node.js, then using npm to install the serverless package globally and verify with serverless -v.

Define your aws deployment with serverless.yaml by setting the service name, provider node.js 22, and a hello function with an http api trigger. CloudFormation uses these instructions to create resources.

Define an AWS DynamoDB table named orders using infrastructure as code in serverless.yaml, with order ID as the partition key to store and manage the bot orders.

Deploy infrastructure as code to AWS using the serverless framework, specify the region, and generate a CloudFormation template to create the orders DynamoDB table, then verify in the AWS console.

Learn to manage Lex conversation flow by implementing a delegate function that keeps asking for missing slots, copies current intent details from the event, and tracks session state and progress.

Implement a place order workflow by using a DynamoDB client and put item command in a Lambda Lex integration, extracting item id and quantity, validating slots, and confirming the order.

Create your first AWS Lambda handler to route Lex intents for the order bot, extract item ID and quantity, delegate for missing slots, and prepare DynamoDB order details.

Define the function in serverless.yaml so Lex can invoke it. Remove unused files, set orderBotHandler in handlers, and deploy with sls deploy.

Grant lambda permission to put items in the orders DynamoDB table using the serverless framework IAM role, deploy, and test with Lex, then troubleshoot with CloudWatch logs to resolve errors.

Create the track order intent, add the order ID slot (amazon.alphanumeric), configure the initial response, decline response, and closing response, and prepare lambda-based fulfillment with tests.

Implement a track order function that queries a DynamoDB orders table with GetItemCommand, extracting the order ID and returning a status-driven detail response, handling missing IDs and not-found results.

Route the request to trackOrder() by updating order.unload.js, importing the trackOrder() intent, then check if intentName equals trackOrder() to return the weight and call trackOrder() with the event, simsql.

Grant Lambda permission to perform DynamoDB:GetItem by updating the serverless.yml, adding the DynamoDB:GetItem action to the IAM role statement.

Apply CloudFormation intrinsic functions to dynamically retrieve the orders table ARN, replacing manual ARN references. Deploy the updated service to AWS and verify the deployment command in the next lecture.

Test the track order intent in AWS Lex by configuring Lambda fulfillment and using DynamoDB orders to simulate status updates, including delivered.

Create an Amazon Kendra indents to store, understand, and search your documents, then configure an IAM role for access and choose the developer edition.

Add an Amazon Kendra data source from Amazon S3, name it my data source S3, configure permissions, select the S3 location, set daily or monthly sync, and test the results.

Create a FAQ intent in Amazon Lex by adding an empty intent, supplying sample utterances like how do I track my order, and linking fulfillment to a lambda function.

Wire Lex to Kendra by initializing a Kendra client, querying with the user question, returning a Lex close response with the best result, and handling errors.

Route requests to the faq such Kendra function by exporting and importing the module, and implement an intent check for faq to return the faq such Kendra response.

Grant the lambda function permission to query amazon kendra by configuring the arn, region, and index id, then deploy the application using the serverless workflow.

Learn to deliver a direct answer in an AWS AI customer support chatbot by guiding users to track their order with an order ID and using Kendra results.

Explore how a voice interaction enables customers to place an order by providing an item ID and quantity, then records the order in DynamoDB and tracks its processing status.

Explore Amazon Comprehend, an AWS natural language processing service that reads text, understands meaning, detects sentiment, and identifies topics and entities to train a classifier routing tickets into business categories.

Discover the building blocks of Amazon Comprehend, including input test, preview models, custom models, and endpoint, and how they turn unstructured tests into insights like sentiment, entities, keyphrases, and language.

Create a DynamoDB table named support tickets to store tickets using infrastructure as code in a serverless framework, duplicating a resource and setting the ticket ID as the partition key.

Deploy the order service with SLS deploy, then verify the DynamoDB tables, including the support tickets table and ticket ID, noting there are currently no items.

Upload customer support datasets to Amazon S3 and reference them with Amazon Comprehend to build a production data source for custom test classification models, like billing and technical.

Create a custom classification model in Amazon Comprehend using a CSV training dataset with two labels, billing and technical, in single-label mode.

Discover what a model is—the brain that learns patterns from data—and understand what it means to be trained, i.e., successfully teaching the brain to recognize data patterns.

Create an endpoint in Amazon Comprehend to connect the category classifier model to an application, provide one inference unit for real-time predictions, and delete it after testing.

Classify customer issues with Amazon Comprehend using the classify document command, determine billing or technical categories, and create tickets stored in DynamoDB with Lex-derived descriptions and a unique ticket id.

Grant a function permission to write data to the DynamoDB support tickets table, configuring a DynamoDB put item action with the correct ARN and CloudFormation usage.

Direct conversations to the createTicket intent by exporting and importing the createTicket handler in orderBotAdler.js, checking the intent name, and returning the createTicket function to handle events.

Go to the AWS Management Console, open Amazon Comprehend, select endpoints, choose actions and delete, then refresh to confirm the endpoints are deleted.