Development guide

Development guide

Setup

Install both nodejs and python on your computer.

Install CDK:

npm install -g aws-cdk

Set up a virtual env:

python3 -m venv .venv

source .venv/bin/activate

find . -name requirements.txt | xargs -I{} pip install -r {}
find . -name requirements-dev.txt | xargs -I{} pip install -r {}

After this initial setup, you only need to run source .venv/bin/activate to use the virtual env for further development.

Deploy the technique examples (AWS Step Functions)

Deploy all the stacks:

cd techniques_step_functions

cdk deploy --app 'python3 technique_stacks.py' --all

Run an example input through each flow example. Edit the files in the test-inputs directory to change the test execution inputs.

./run-test-execution.sh ModelInvocation

./run-test-execution.sh PromptTemplating

./run-test-execution.sh SequentialChain

./run-test-execution.sh ParallelChain

./run-test-execution.sh ConditionalChain

./run-test-execution.sh HumanInput

./run-test-execution.sh Map

./run-test-execution.sh AwsServiceInvocation

./run-test-execution.sh Validation

Deploy the technique examples (Amazon Bedrock Flows)

Deploy all the stacks:

cd techniques_bedrock_flows

cdk deploy --app 'python3 technique_stacks.py' --all

Run an example input through each flow example. Edit the files in the test-inputs directory to change the test execution inputs.

python3 run-test-execution.py ModelInvocation

python3 run-test-execution.py PromptTemplating

python3 run-test-execution.py SequentialChain

python3 run-test-execution.py ParallelChain

python3 run-test-execution.py ConditionalChain

python3 run-test-execution.py HumanInput

python3 run-test-execution.py Map

python3 run-test-execution.py AwsServiceInvocation

python3 run-test-execution.py Validation

Deploy the demo application

Fork this repo to your own GitHub account. Edit the file cdk_stacks.py. Search for parent_domain and fill in your own DNS domain, such as my-domain.com. The demo application will be hosted at https://bedrock-serverless-prompt-chaining.my-domain.com. Push this change to your fork repository.

Create a new GitHub personal access token. The token only needs the public_repo scope. Copy the generated token, and create a Secrets Manager secret containing the token:

aws secretsmanager create-secret \
    --region us-west-2 \
    --name BedrockPromptChainGitHubToken \
    --description "For access to public repos for the Bedrock serverless prompt chain demos" \
    --secret-string "{\"token\": \"<your token>\"}"

Set up a Weasyprint Lambda layer in your account. One of the examples in the demo application uses this library to generate PDF files.

wget https://github.com/kotify/cloud-print-utils/releases/download/weasyprint-63.0/weasyprint-layer-python3.13.zip

aws lambda publish-layer-version \
    --region us-west-2 \
    --layer-name weasyprint \
    --zip-file fileb://weasyprint-layer-python3.13.zip \
    --compatible-runtimes "python3.13" \
    --license-info "MIT" \
    --description "fonts and libs required by weasyprint"

aws ssm put-parameter --region us-west-2 \
    --name WeasyprintLambdaLayer \
    --type String \
    --value <value of LayerVersionArn from above command's output>

Deploy all the demo stacks:

cdk deploy --app 'python3 cdk_stacks.py' --all

The demo application will be hosted at https://bedrock-serverless-prompt-chaining.my-domain.com, behind Cognito-based user authentication. To add users that can log into the demo application, select the bedrock-serverless-prompt-chaining-demo user pool on the Cognito console and click "Create user".

As part of deploying the demo application, an SNS topic bedrock-serverless-prompt-chaining-notifications will be created and will receive notifications about demo failures. An email address or a chat bot can be subscribed to the topic to receive notifications when the demo's alarms fire.

Deploy the demo pipeline

The demo pipeline will automatically keep your deployed demo application in sync with the latest changes in your fork repository.

Edit the file pipeline/pipeline_stack.py. Search for owner and fill in the GitHub account that owns your fork repository. Push this change to your fork.

Deploy the pipeline:

cdk deploy --app 'python3 pipeline_app.py'

Activate the CodeStar Connections connection created by the pipeline stack. Go to the CodeStar Connections console, select the bedrock-prompt-chain-repo connection, and click "Update pending connection". Then follow the prompts to connect your GitHub account and repos to AWS. When finished, the bedrock-prompt-chain-repo connection should have the "Available" status.

Go to the pipeline's page in the CodePipeline console, and click "Release change" to restart the pipeline.

As part of deploying the demo application, an SNS topic bedrock-serverless-prompt-chaining-notifications will be created and will receive notifications about pipeline failures. An email address or a chat bot can be subscribed to the topic to receive notifications when pipeline executions fail.

Test local changes

Ensure the CDK code compiles:

cdk synth

Run the webapp locally:

docker compose up --build

Changes to Step Functions state machines and Lambda functions can be tested in the cloud using cdk watch, after the demo application has been fully deployed to an AWS account (following the instructions above):

cdk watch --app 'python3 cdk_stacks.py' \
    PromptChaining-BlogPostDemo \
    PromptChaining-TripPlannerDemo \
    PromptChaining-StoryWriterDemo \
    PromptChaining-MoviePitchDemo \
    PromptChaining-MealPlannerDemo \
    PromptChaining-MostPopularRepoBedrockAgentsDemo \
    PromptChaining-MostPopularRepoStrandsDemo

Then in a separate terminal, run test executions in the cloud after making changes to your code. Edit the files in the test-inputs directory to change the test execution inputs.

./run-test-execution.sh BlogPost

./run-test-execution.sh TripPlanner

./run-test-execution.sh StoryWriter

./run-test-execution.sh MoviePitch

./run-test-execution.sh MealPlanner

./run-test-execution.sh MostPopularRepoBedrockAgents

./run-test-execution.sh MostPopularRepoStrands