Very new to this concept and programming, but really started trying to learn after watching Amazons AI predict blitzes and want to try to do that myself to challenge and learn something new.
If I were you I would start by searching online information about the models Amazon uses for this specific use case. I would then break down the problem into pieces and learn related concepts in all relevant directions (programming, ML, … ). Some relevant ML concepts/techniques/tools can be object detection, object tracking, graph machine learning, attention, etc…
It looks like there is not a lot of information available on the internet as we speak. Most insighful i could find out was here, and it is quite high-level:
With the Defensive Targets feature, Amazon Prime has trained machine learning models on 35,000+ plays to guess when it’s going to happen. Body language, pre-snap position, and general player movements are all taken into account to provide viewers with potentially game-changing plays in real time.
Also, if I may advise -based on my past academic research- set ambitious goals but be realistic at the same time: you will need data, and you will need computing power.