Use Generative AI To Create Live Sports Commentary

The demo is built with AWS services focused in 3 workflows: 1. Live game telemetry data ingestion, 2. Perform Generative AI model inference, 3. Presents AI Generated commentary in the client. When a live event occurs, such as a play in American football game, or a turn over in a basketball game, the data that described the play is produced in realtime in continuous fashion. In order to ingest these data as soon as it becomes available, a streaming solution is needed to capture all the actions for downstream processing. We choose Kinesis data stream as the streaming solution in our demo. The number of Kinesis streams/shards needed to support the solution is determined by how frequent an event is created from the source, and how many events are to be captured simultaneously. Once the telemetry data is ingested into a Kinesis data stream, a lambda function gets triggered. It gathers and analyzes the input and to turn the telemetry data into a prompt to be used by the GenAI model for a commentary. In this demo, we use Jurassic-2 Ultra model hosted in Amazon Bedrock for creating the live commentary. Once a commentary has been generated, the lambda function pushes the data into a target Kinesis data stream for downstream consumers, such as a web client or application for presentation purposes. We design a target Kinesis data stream to allow any additional downstream application which might be subscribing to the generated commentary, such as persisting to a DynamoDB, or S3 bucket so that the commentaries could be replayed at later time, opportunity for reselling purposes or simply for archival.