Hi there,
OK, so slow down with that…
I have been following this thread and I don’t think it’s been sufficiently diagnosed and tested to make such a stark determination.
“pretty but not supported”
You are NOT the first customer of this SO understand 2 things. It’s new tech and NOT all the functions are understood, the info available needs updating, more changes are in the process as well as some WiKi updates. Be patient and give them time to deduce the solution for your issue. Meanwhile can you Say it detects people accurately ? and can it decerne with HIGH confidence from a bicycle or a dog ? How did you train the model ? Did you train it?
I can tell you from experience’s with both , the Unknown category needs 3 times as many samples, Boosts the confidence of known people’s.
Can I ask , Does the model detection work? before you add in the LORA portion and the whole TTN business ( I frankly don’t see the point of it? ) Probably because I’m not understanding the Whole purpose.
So there are a lot of smart folks on here willing to help be patient and they will chime in more. AFAIK there are almost NO complaints about the quality of the tech and operation so , hang in there. You’ll get it going.
Can you post your question here so we can analyze please?
HTH
GL 
PJ 
IMO, it’s a good platform and great hardware , but there is a learning curve with all it can do, it crosses over into many areas of tech. 