Summarization Plugin

Note

The AWS Support Cases Summarization powered by Amazon Bedrock may not capture all nuances of the original conversation with the AWS Support. It should be verified against the full transcript which remains the single source of truth for accuracy and completeness. See AWS Responsible AI Policy.

Note

The AWS Support Cases Dashboard is not realtime and might have 24h-48h delay. For the ongoing cases please check AWS Support Center in the respective account.

Authors

• Samuel Chniber, Senior Solution Architect

• Iakov Gan, Senior Solution Architect

• Yuriy Prykhodko, Principal Technical Account Manager

Feedback & Support

Follow Feedback & Support guide

Introduction

AWS Support Cases Summarization is a Plugin to the AWS Support Cases Radar Dashboard that leverages the power of Generative AI (GenAI) through the use of Amazon Bedrock.

This plugin aims at summarizing the AWS support cases problem statement, communications with the AWS Support Engineers and recapping eventual actions to be carried out either by AWS or by the customer for resolution.

Architecture Overview

The Summarization Stack deploys a rule to Default Eventbridge bus to capture events sent by the Data Collection Stack. The Eventbridge Rule processes the message by sending it to an Amazon SQS queue. The SQS Queue is responsible for triggering a lambda function that executes Bedrock API call for the summarization and enriches the collected support case data by adding the summaries back to Amazon S3.

Deployment steps

Step 1 of 4: Deploy the AWS Support Cases Radar Dashboard

This plugin has a dependency on the successful deployment of the AWS Support Cases Radard Dashboard.

Read More

Step 2 of 4: Enable Foundation Model on Amazon Bedrock

To get AWS Support Cases Summarized you need to add access to Amazon Bedrock foundation models.

Click here to see instructions on the AWS Console

Step 3 of 4: (Optional) Deploy an Amazon Bedrock Guardrail in the Data Collection Account in the Inference Region

Amazon Bedrock Guardrails is a crucial security feature for generative AI applications that helps implement safeguards based on specific use cases and responsible AI policies. It provides an additional layer of protection on top of the native safeguards offered by foundation models (FMs).

We provide an example of Amazon Guardrails stack, but if your company is already using Guardrails you can skip this Step and continue to installation of the Plugin Stack (Step 4).

Click here to know more about plugin stack parameters

This plugin comes with the following reasonable defaults that can be overridden through the parameters exposed by the CloudFormation template:

Parameter	Description	Default
BlockedInputMessage	Message to return when the Amazon Bedrock Guardrail blocks a prompt.	{"executive_summary":"Amazon Bedrock Guardrails has blocked the AWS Support Case Summarization.","proposed_solutions":"","actions":"","references":[],"tam_involved":"","feedback":""}
BlockedOutputMessage	Message to return when the Amazon Bedrock Guardrail blocks a model response	’’
IncludeSexualContentFilter	Whether to include Sexual Content Filter in the Guardrail or not	'yes'
SexualContentFilterInputStrength	The strength of the content filter to apply to prompts. As you increase the filter strength, the likelihood of filtering harmful content increases and the probability of seeing harmful content in your application reduces.	'HIGH'
SexualContentFilterOutputStrength	The strength of the content filter to apply to model responses. As you increase the filter strength, the likelihood of filtering harmful content increases and the probability of seeing harmful content in your application reduces	'HIGH'
IncludeViolentContentFilter	Whether to include Violent Content Filter in the Guardrail or not	'yes'
ViolentContentFilterInputStrength	The strength of the content filter to apply to prompts. As you increase the filter strength, the likelihood of filtering harmful content increases and the probability of seeing harmful content in your application reduces	'HIGH'
ViolentContentFilterOutputStrength	The strength of the content filter to apply to model responses. As you increase the filter strength, the likelihood of filtering harmful content increases and the probability of seeing harmful content in your application reduces	'HIGH'
IncludeHateContentFilter	Whether to include Violent Content Filter in the Guardrail or not	'yes'
HateContentFilterInputStrength	The strength of the content filter to apply to prompts. As you increase the filter strength, the likelihood of filtering harmful content increases and the probability of seeing harmful content in your application reduces	'HIGH'
HateContentFilterOutputStrength	The strength of the content filter to apply to prompts. As you increase the filter strength, the likelihood of filtering harmful content increases and the probability of seeing harmful content in your application reduces	'HIGH'
IncludeInsultsContentFilter	Whether to include Insults Content Filter in the Guardrail or not	'yes'
InsultsContentFilterInputStrength	The strength of the content filter to apply to prompts. As you increase the filter strength, the likelihood of filtering harmful content increases and the probability of seeing harmful content in your application reduces	'HIGH'
InsultsContentFilterOutputStrength	The strength of the content filter to apply to prompts. As you increase the filter strength, the likelihood of filtering harmful content increases and the probability of seeing harmful content in your application reduces	'HIGH'
IncludeMisconductContentFilter	Whether to include Insults Content Filter in the Guardrail or not	'yes'
MisconductContentFilterInputStrength	The strength of the content filter to apply to prompts. As you increase the filter strength, the likelihood of filtering harmful content increases and the probability of seeing harmful content in your application reduces	'HIGH'
MisconductContentFilterOutputStrength	The strength of the content filter to apply to prompts. As you increase the filter strength, the likelihood of filtering harmful content increases and the probability of seeing harmful content in your application reduces	'HIGH'
IncludePromptAttackContentFilter	Whether to include Insults Content Filter in the Guardrail or not	'yes'
PromptAttackContentFilterInputStrength	The strength of the content filter to apply to prompts. As you increase the filter strength, the likelihood of filtering harmful content increases and the probability of seeing harmful content in your application reduces	'HIGH'

Step 3 of 4: Deploy the AWS Support Case Summarization Stack In the Data Collection Account

In this step we will deploy the summarization Plugin stack via cloud formation.

Click here to know more about plugin stack parameters

This plugin comes with the following reasonable defaults that can be overridden through the parameters exposed by the CloudFormation template:

Parameter	Description	Default
BedrockRegion	The AWS Region from which the Summarization is performed	us-east-1
Instructions	Additional instructions passed to the Large Language Model for the summarization process customizability	’’
Provider	Large Language Model Provider for the summarization process customizability	Anthropic
FoundationModel	Foundation Model to be used for the summarization process	Claude 3.5 Sonnet
InferenceType	Summarization process Inference Type	"ON_DEMAND"
Temperature	Summarization process Temperature	0
MaxTokens	Summarization process Maximum Tokens	8096
MaxRetries	Summarization process Maximum Retries	30
Timeout	Summarization process Timeout in seconds	60
BatchSize	Summarization process Batch Size for parallel processing	1
GuardrailId	Amazon Bedrock Guardrail ID to be used (Use this parameter in case you are using an externally managed Guardrail configuration. Leave empty if not planning to use Amazon Bedrock Guardrail)	’’
GuardrailVersion	Amazon Bedrock Guardrail Version to be used (Use this parameter in case you are using an externally managed Guardrail configuration. Leave empty if not planning to use Amazon Bedrock Guardrail)	’’
GuardrailTrace	The trace behavior for the guardrail	"ENABLED"

Post Deployment

Where i can see summarizations?

You will be able to see summarizations in the dashboard. There are short executive summaries in the table with cases and also a more detailed information with the summary of solution and current actions in the table below once you select the case. The summaries only appear after 48-24h and only for updated cases.

Can I force summarization for all cases?

Yes, you can trigger the refresh of all Support Cases and it will generate summaries as well.

1. Go to the Amazon S3 Bucket of data collection and delete the folder support-cases/ to trigger collection for the last 12 months. (Or you can modify last_read field in json file s3://cid-data-XXX/support-cases/support-cases-status/payer_id=XXX/XXX.json to trigger new collection and summaries generation for less then for 12 months).

2. Go to Amazon Step Functions and execute CID-DC-support-cases-StateMachine this must collect all Support Cases.

3. You can check the activity of the lambda CID-DC-support-case-summarization-Lambda. Make sure this lambda is not triggered errors (Typical issue is the enabling of access to the model. See above). 1. Go to QuickSight and refresh the dataset support_cases_communications_view. Once the refresh finishes you can check the Dashboard. The table with cases and case details must contain executive and detailed summaries.

How often Summarization are updates?

The summarization plugin automatically generates summaries of support case histories based on past communications. The data collection by default operates on a daily basis. Once the plugin is deployed, the summarization stack enriches cases with summaries right after daily collecting new data. Each day, data collection gathers all communications that have occurred since the previous collection. The QuickSight dashboard updates data on the daily schedule as well. As a result, there may be up to a 24-hour delay between when a communication occurs and when it appears in the dashboard. We recommend using the AWS Support center to get the latest information on the status of the current cases.