AWS AI Camera Streaming PoC

I need a concise proof-of-concept that shows an end-to-end pipeline for real-time video analytics on AWS. Three LeekGoVision AI cameras will stream live footage into the cloud. From there the flow should look like this:

1. Video ingestion – push each RTSP stream into AWS (Kinesis Video Streams or another managed entry point).
2. Processing & annotation – invoke AWS Rekognition in real time to run object and facial recognition. Detected entities must be overlaid or otherwise marked inside the stream so I can see live annotations.
3. Storage – archive the raw video in S3 with sensible lifecycle rules; write every annotation (with bounding-box data, confidence score, timestamp, camera ID, and short text label) to a relational table. A small RDS instance is fine for the PoC.
4. Retrieval layer – build a lightweight API or dashboard where I can filter by time range, camera, person, or object and instantly pull up the corresponding video clip or annotated frame.

The solution must stay almost entirely inside AWS managed services; custom code should be limited to glue logic, Lambdas, or a slim front-end. I will supply the AWS account and camera credentials—you wire everything together, document the setup steps, and leave me with:

• CloudFormation/Terraform (or clear CLI commands) to recreate the stack
• Source code for any custom functions or UI components
• A short README describing how to plug in additional cameras and expand the schema

If the PoC reliably detects objects and faces, saves the annotations, and lets me retrieve a clip in under five seconds, we are good.