Multi-interface chatbot using Amazon Q and Slack with CloudFront clickable sources

Introduction:

This application is an AI-powered chatbot designed to help users navigate and utilize the Spack package manager tool. This innovative chatbot that employs Retrieval-Augmented Generation (RAG) technology to provide responses to user queries and can be interacted with both in Slack and Amazon Q. This solution can be applied to a variety of use cases that require knowledge integration to a chatbot. The two main selling points of this solution is: (1) having the ability to interact with the chatbot in two mediums and (2) being able to show sources in websites and documents in an S3 bucket via CloudFront.

Notable Features:

• Slackbot and Amazon Q both share the same vector database reducing solution complexity and cost.
• Both Slackbot and Amazon Q bot provides clickable links to its sources for generating answers in their responses:
  • https://spack-tutorial.readthedocs.io/en/latest/index.html
  • Text files (docuemnts/slack/*.txt) uploaded to S3 bucket via CloudFront distribution
• Automated updates of both data sources.
• Dashboard for tracking bot invocations for Slack and Amazon Q:
  • Aws Console -> Amazon Q Business -> Radiuss -> Analytics dashboard
  • Aws Console -> Cloudformation -> Stacks -> SlackStack -> Outputs -> AmazonQCloudwatchDashboardOutput

Architecture:

Stacks:

• Data Stack:

  • A) Documentation Data
    1. Documentation Processing Lambda pulls in data from Raw Documentation Bucket and does the following:
       • Converts .rst files into markdown.
       • Splits the markdown text based on its title
       • Generates metadata files that will be used by kendra. The metadata files contains the following attributes:
         • title: section title from data split
         • data_source: documentation
         • _source_uri: URL from the documentation which is https://spack.readthedocs.io/en/latest/ + file name + "#" + section title
    2. Documentation Processing Lambda saves the split markdown and the metadata files into Processed Documentation Bucket.
    3. Documentation Processing Lambda triggers a kendra data source sync job to crawl the Processed Documentation Bucket.
  • B) Slack Data Slack Processing Lambda:
    1. Slack Processing Lambda pulls in data from Raw Slack Bucket which contains historical Slack data and does the following:
       • Generates metadata files that will be used by kendra. The metadata files contains the following attributes:
         • title: section title from data split
         • data_source: slack
         • _source_uri: generated CloudFront URL from the Raw Slack Bucket
    2. Slack Processing Lambda saves historical slack data and the metadata files into Processed Slack Bucket.
    3. Slack Processing Lambda triggers a kendra data source sync job to crawl the Processed Slack Bucket.
    4. Raw Slack Bucket data is passed into a CloudFront distribution for public access. Slack Ingest Lambda:
    5. Slack Ingest Lambda is triggered by event bridge daily.
    6. Slack Ingest Lambda pulls in the past 24 hours conversation from slackdata from slack and writes it to Raw slack data
    7. Slack Ingest Lambda saves conversation data into Processed Slack Bucket together with its metadata.
    8. Slack Processing Lambda triggers a kendra data source sync job to crawl the Processed Slack Bucket.
    9. Processed Slack Bucket data is passed into a CloudFront distribution for public access.

• Amazon Q Stack: Amazon Q Business is a fully managed, generative-AI powered assistant tailored for this use case to answer questions based on the data from the data stack.

1. Identity Center - Provides authentication to Amazon Q.
2. Kendra provides context to the responses via semantic search and sources.
3. Cloudfront links are provided by kendra and propagates to the responses of Amazon Q.
4. Public Docs links are provided by kendra and propagates to the responses of Amazon Q.
5. Invocations are logged in a Cloudwatch Dashboard.

• Slack
  • A) Answering Questions
    1. Slack app invokes API Gateway with the question as a part of the payload.
    2. API Gateway invokes Slackbot Lambda.
    3. Slackbot Lambda pulls Slack token from Secrets Manager.
    4. Slackbot Lambda pulls Slack parameters for responses from SSM Parameter Store.
    5. Kendra is queried with the question and responds with relevant passages and sources from documentation and slack data from Cloudfront.
    6. Public docs are returned as part of the response if the chatbot used it as a source.
    7. Slack data via Cloudfront are returned as part of the response if the chatbot used it as a source.
  • B) Reporting 0. Metrics Lambda is triggered every day at 0:00 UTC
    1. Everytime the Slackbot Lambda is triggered it is captured in Cloudwatch as a metric.
    2. Metrics Lambda pulls daily data from Cloudwatch
    3. Metrics Lambda pulls Slack token from Secrets Manager
    4. Metrics Lambda pulls slack parameters for responses from SSM Parameter Store
    5. Metrics Lambda send message on slack with daily report

Deploying the Solution:

Requirements:

• Active AWS account
• Docker
• AWS CLI
• Slack workspace
  • Parent Channel: Public Slack Channel where users will be interacting with the slack chatbot.
  • Child Channel: Private Slack Channel where metrics report will be sent to.
• If building from an arm based machine (Apple M series) change the parameter for architecutre in documentation_processing_lambda in stacks/data.py to lambda_.Architecture.ARM_64.

Step 1: Account Set-up

• Ensure Bedrock model access for anthropic.claude-v2:1
• Enable IAM identify Center for your account.

Step 2: Install project dependencies

pip install -r requirements.txt

Step 3: Bootstrap Account (Skip if done previously)

cdk bootstrap

Step 4: Synthesize CDK App

cdk synth

Step 5: Deploy AWS infrastructure

cdk deploy --all

Step 6: Slack Set-up

1. Create a Slack app:

• Go to: https://api.slack.com/apps
• Select Create an App
• Select From a manifest
  1. Select Spack workspace
  2. Select YAML tab and copy the contents of the app manifest and select Next
  3. Select Create

2. Install App

• On the left pane, under settings select Install App
• Select Install to <workspace>
• Select Allow

3. Copy Bot User OAuth Token

• Go to: AWS Console -> AWS Secrets Manager -> Secrets -> SlackAccessKey### -> Overview -> Retrieve Secret Value -> Edit
• Paste value where it says place-holder-access-key
• Click Save

4. Enter endpoint (One app is finished deploying from Step 1)

• Got to: AWS Console -> Cloudformation -> Stacks -> SlackStack -> Outputs -> SlackBotEndpointOutput (copy Value)
• Enable events

• Paste value under Request URL

• On the bottom right of the screen select Save Changes

5. Invite bot to channels (Parent and Child):

• Select the channel
• On the upper right next to huddle click on the three dots.
• Select edit settings
• Go to integrations tab
• Select Add an app
• Under the In your workspace tab select add the chatbot

6. Enter Slack workspace information:

Obtain the following information from slack:

• Parent channel ID
• Child channel ID

Note: To obtain channel id, right-click the channel -> View Channel Details -> About -> Copy channel ID

• Slackbot member ID: Under apps -> right-click the bot -> view app details -> Copy Member ID. If Slack bot is not under apps, click Add apps and select the slackbot.

Enter above information into AWS:

• Got to: AWS Console -> Systems Manager -> Application Management -> Parameter Store -> My parameters
• Select /Radiuss/Spack/ChildChannelId and edit. Enter the child channel id as the value and select save changes
• Select /Radiuss/Spack/ParentChannelId and edit. Enter the parent channel id as the value and select save changes
• Select /Radiuss/Spack/SlackbotMemberId and edit. Enter the Slackbot member id as the value and select save changes

Security: It is highly recommended that the user change the slack token periodically.

Step 7: Adding users to Amazon Q

• Go to AWS Console -> Amazon Q Business -> Applications -> Radiuss -> User Access -> Manage user access
• Select Add groups and users
• Select Add and assign new users
• Select Next
• Enter information
• Select Next
• Select Add

Upgrade/Downgrade user subscription

• Go to AWS Console -> Amazon Q Business -> Applications -> Radiuss -> User Access -> Manage user access
• Select user via radio button
• Select Edit subscription
• Select Choose subscription from dropdown Subscription tiers are available in this link

Accessing the applications:

• Amazon Q: AWS Console -> Amazon Q Business -> Applications -> Radiuss -> Web experience settings -> Deployed URL
• Slack: Workspace -> Designated Channel -> Send a single message that starts with @SpackChatbot

Clean up:

cdk destroy --all

Authors and Reviewers:

• Nick Biso, Machine Learning Engineer - Amazon Web Services Inc.
• Ian Lunsford, Aerospace Cloud Consultant - Amazon Web Services Inc.
• Natasha Tchir, Machine Learning Engineer - Amazon Web Services Inc.
• Katherine Feng, Machine Learning Engineer - Amazon Web Services Inc.

Security

See CONTRIBUTING for more information.

License

This library is licensed under the MIT-0 License. See the LICENSE file.