I do like the onboard AI, and it works well for entity detection (like people). We haven't found the face detection to be very reliable in outdoor security applications. There doesn't seem to be a way to correct/combine classes if someone's detected as multiple individuals on different occasions, so we end up with the same person detected as 5 "unknown"s. This is not a hard problem to solve. You'd just allow embedding matching to different face groups, but it's annoying as a user.