YOLOv8n Inference Engine Development

I'm looking for a professional to help me develop an inference engine for analytic processing of open YOLOv8n models. The objective is to detect persons and vehicles from the YOLOv8n model and process RTSP streams.

Key Requirements:
I want to develop a inference engine for analytic processing of open yolo models,
1. i want to detect person and vehicles from yolo v8n model
2. i want to process rtsp stream
3. i want to process each rtsp at 2fps only
4. there should be 50 rtsp stream i need to process on single server Intel Xeon 24 core with Nvidia A2 GPU.
5. processing frame size should be 640 only
6. i want to detect person and vehicle on every camera and need alert in json
7. it should be real time
8. message queue protocol should be implemented like kafka or Zmq

Although I have skipped some questions, I am open to suggestions on the main application area for this inference engine, the preferred type of output from the inference engine, and any specific conditions or environments for the RTSP streams. Your expertise could greatly help in defining these aspects.