Dynamic FAST channels powered by Amazon Bedrock

In today’s streaming landscape, personalization and an enhanced user experience are paramount. This walk-through proposes a dynamic Free Ad-supported Streaming Television (FAST) channels solution integrated with generative AI to enrich the user experience. This solution achieves personalization based on targets, segments and any other information related to the user. By delivering personalized content, FAST channels can create an immersive and engaging experience for viewers, driving content consumption, user retention, and growth by as much as 25%.

About generative AI for FAST channel creation

Amazon Bedrock is a fully managed service that offers a choice of high-performing foundation models (FMs) from leading AI companies including AI21 Labs, Anthropic, Cohere, Meta, Mistral AI, Stability AI, and Amazon. It simplifies the development of generative AI applications while maintaining privacy and security.

With Amazon Bedrock, you can:

Experiment with various top FMs
Customize models privately using techniques like fine-tuning and Retrieval-Augmented Generation (RAG)
Create managed agents for complex business tasks without writing code
Securely integrate and deploy generative AI capabilities using familiar AWS services

Personalized dynamic FAST channels engage audiences, enhance the user experience, and create optimal targeting for business objectives. Amazon Bedrock with Anthropic Claude Sonnet 3.5 model enables live personalization through interaction with users, providing continuous feedback for content and advertising recommendations.

In this blog post, we explain how to create a dynamic FAST channel (based on personalization leveraging AI generative models) to generate a channel with contextually relevant content that is tailored to the user’s profile.

We will use Amazon Web Services (AWS) serverless technology, which includes:

Amazon Bedrock to invoke large language models (LLM) for content generation
Amazon DynamoDB tables to store user information and data
AWS Lambda integration
AWS Fargate with Amazon API Gateway to expose APIs to the web application

Overview of the solution

Figure 1: Demo reference architecture and sequence of steps.

The steps for this workflow are as follows:

First AWS Data Exchange provides the IMDb dataset, creating the source for content search and access, an essential part needed for Amazon Bedrock with Claude Sonnet 3.5 to make content recommendations. The Intel Xe Super Sampling (XeSS) process triggers the start of the flow where it is deposited in an Amazon Simple Storage Service (Amazon S3) bucket as a result of the processing. (This procedure is optional, since any material can be uploaded and not necessarily all materials go through the XeSS process.)
The key piece of FAST channel personalization in this solution is the user information (such as tastes, content preferences, as well as the history of content played). All this information is stored in an Amazon DynamoDB database and then Amazon Bedrock with Claude Sonnet 3.5 uses this information to generate a personalized FAST channel. There is an additional key source of information, the conversational agent, which allows the use of prompts, such as “Can you recommend a FAST channel for action and adventure movies?” Amazon Bedrock uses the prompt to return a channel with content recommendations from the IMDb database.

Figure 2: The user profile contains relevant information such as genre preference and history of content viewed.

AWS Lambda converts the comma-separated values (CSV) dataset of movie information into the required JSON format. This script incorporates several key features: it ensures data integrity by validating mandatory fields (title, titleId, duration, rating, and image_url), handles complex data types like lists and comma-separated strings, and optimizes the structure of AMAZON_BEDROCK_METADATA and AMAZON_BEDROCK_TEXT_CHUNK for improved search relevance. The resulting output is a series of well-structured JSONL files, each containing 100 records, which are optimized for ingestion into Amazon Bedrock Knowledge Bases. This solution not only prepares your data for efficient querying and retrieval but also provides a foundation for maintaining data quality and facilitating future updates to your knowledge base. The result is a streamlined data pipeline that transforms the streaming service’s movie dataset into a format perfectly tailored for Amazon Bedrock Knowledge Bases. This optimization enables more accurate and relevant movie recommendations, enhancing the user experience and potentially increasing viewer engagement by as much as 25%.
Once the content is uploaded in the S3 bucket, AWS Elemental MediaConvert automatically takes the asset and processes it to convert it into HTTP live streaming (HLS) format. This HLS output is deposited in the S3 bucket and is taken by AWS Elemental MediaPackage to package it in Dynamic Adaptive Streaming over HTTP (DASH) and common media application formats (CMAFs). The whole process is automatic through AWS Lambda.
AWS Elemental MediaTailor takes the assets already packaged from MediaPackage as a result of the Amazon Bedrock transaction and adds them as a source location. It creates the channels within its Channel Assembly section, adding the assets as channel programs. It also adds the commercial insertion with the Ad Insertion section of AWS Elemental MediaTailor and the integration with the Ad Decision server. This process is also automated through the use of AWS Lambda.
AWS Elemental MediaTailor Channel Assembly exports the schedule to the Amazon DynamoDB database to feed the Planby (Open source) EPG for the custom FAST channel that has been created.

Figure 3: The channel created consists of a personalized list for each user with its respective EPG.

The user receives the personalized FAST channel based on the user’s profile, content history or conversational agent prompt. In the web interface, you can see the channel in HLS, DASH and CMAF format distributed by Amazon CloudFront with its EPG—the user must be logged in. The application was developed with AWS Amplify, which allows the use of a full-suite of tools and services specifically designed to help developers easily build and launch apps.

Figure 4: User interaction with the Amazon Bedrock agent.

Conclusion

This blog post describes a solution that generates audience engagement through the use of FAST channels based on end-user preferences, behavior and interactions. When created in conjunction with generative AI, these channels can create value-added content recommendations and personalization—boosting engagement by as much as 25%.

By integrating Amazon Bedrock to invoke large language models with a serverless architecture on AWS, we created an automated pipeline that integrates custom datasets, the IMDb dataset, or any other metadata to produce contextual recommendations tailored to the user profile and user interactions.

Overall, this solution showcases how AWS serverless building blocks can be combined with the unique capabilities of generative AI, available on Amazon Bedrock using Claude Sonnet 3.5, to transform viewer metadata for smart content personalization.

Contact an AWS Representative to learn how we can help accelerate your business.