/namer-hackathon

Primary LanguagePython

namer-hackathon

Generate a Podcast from the AWS whats new RSS Feed

This repo contains an AWS SAM definition and a sample streamlit app to play your podcast.

Pre-Reqs

  1. Sagemaker Endpoint running Llama-2-7b-chat (tested on ml.g5.2xlarge)
  2. AWS CLI
  3. AWS SAM CLI see https://docs.aws.amazon.com/serverless-application-model/latest/developerguide/install-sam-cli.html
  4. boto3 installed pip install boto3
  5. Note: If you use AWS Cloud9, steps 2,3,and 4 above are already preinstalled.
  6. Streamlit pip install streamlit

Instructions

  1. Navigate to the SAM folder cd SAM
  2. Build the SAM package sam build --use-container
  3. Deploy the package.sam deploy --guided Give your stack a name, and use the region where your SageMaker Endpoint is deployed. Use the defaults for the rest of the options.
  4. The following resources will be deployed: AWS Step Functions State Machine, AWS S3 Bucket, 7 Lambda functions, various IAM roles and policies.
  5. Navigate to the deployed state machine and choose New Execution
  6. Open the included sampleStepFunctionInput.json file, update the following fields, and save:
  • Replace the bucket with the bucket name deployed by the SAM application
  • Replace llmEndpoint with your SageMaker Endpoint Name.
  • Replace numDays with the number of days from today that you want to process from the RSS feed.
  1. Execute the Step Function from the command line: aws stepfunctions start-execution --state-machine-arn <YOUR STATE MACHINE ARN> --input "$(jq -R . sampleStepFunctionInput.json --raw-output)"
  2. Depending on the number of new announcements in the time range that you specified in numDays and your SageMaker Endpoint instance type, execution can take anywhere from 5-15 minutes or longer.
  3. Monitor the execution by running the command: aws stepfunctions describe-execution --execution-arn <YOUR EXECUTION ARN>. Get the Execution ARN from the output of Step 7.
  4. Once complete, grab the runID and bucket name from the output of Step 9. You will need this if you want to run the Streamlit app below.

Listen to your podcast

You have 2 options:

  1. Download the podcast.mp3 file from the s3 bucket
  2. Run the included Streamlit app sampleApp.py

To run the Streamlit app, do the following:

  1. In your terminal, open sampleApp.py
  2. Replace bucketName with your s3 bucket
  3. Replace runId with the UUID you copied earlier
  4. Save and close
  5. Run streamlit run ./sampleApp.py
  6. Open the URL in your browser. If using Cloud9, you must have your EC2 instance in a public subnet with a public IP, and a Security Group rule allow traffic to the Streamlit port (usually 8501)

Cleanup

1.Empty the S3 bucket or you will get an error when you delete the SAM application 2. aws sam delete --stack-name <your stack name here>

Explore the objects in the s3 bucket.

  • processed/ - Contains the metadata and scraped content from each RSS feed item, populated by the processRSS function.
  • dialogs/ - Contains the outputs from the LLM generated by the generateDialog function for each RSS feed item.
  • topic_audio/ - Contains the individual mp3 objects for each RSS feed item, generated by Amazon Polly, from the generateTopicAudio function.
  • transition_audio/ - Contains the individual mp3 objects for the transitions between topic categories, generated by the generateTransitions function.
  • other_audio/ - Contains the individual mp3 objects for the Intro and Outro, generated by the generateIntro and generateOutro functions.
  • playlist.json - Sequenced manifest of each mp3 object. Output by the generatePlaylist function. Used by the Streamlit app to provide titles and links.
  • podcast.mp3 - Final mp3 object combining all individual mp3's, output by the generatePlaylist fucntion.