Implementing Video Intelligence in the Cloud: AI Video Analysis with TwelveLabs Part 2 – Building an Analytics Pipeline Using AWS Media Services in Near-Real-Time Environments | Amazon Web Services

TwelveLabs Pegasus is a video understanding model that interprets scenes and context and returns analysis as text.

An Agent can let you define events in natural language and send near‑real‑time alerts when those events are detected.

AWS Elemental provides managed, broadcast‑quality live video processing; MediaLive can ingest RTMP/RTP/HLS/MP4 sources and write 10‑second HLS segments (.ts and .m3u8) directly to S3.

Each S3 .ts creation triggers EventBridge to invoke a Lambda, which uses FFmpeg to convert .ts to MP4, uploads the MP4 to an analysis S3 bucket, and calls the TwelveLabs API to start analysis.

Segment‑level analysis limits detection of continuous events; MediaPackage time‑shift and a configured time‑shift window enable DVR‑style access to past stream portions.