Just thought I'd mention my approach to this (not 3 phase).
I have had a clamp meter for a number of years (single phase), it was part of a smart setup given away by an electricity company and transmitted the data to a display in the 433 MHz range. I didn't have the display, but read the values using rtl_433 and used that in my home automation system for monitoring.
This was generally OK, but not too accurate - I was a few kWh out a day.
I've recently revisited this, and have set up a pi zero W with a webcam pointing at my meter (non-smart but with numeric display rather than dials) running motioneye OS. Just so I can read the meter without opening up the little cupboard it sits in.
I then set up python opencv on my HA Pi, and wrote a small python script to track digit change and send the reading back to my mqtt server.
I haven't gone full ocr yet as I've found no real need. I simply identify a small gap between digits, crop a small window, greyscale it, threshold it, and then count the number of white pixels in the image - meter is white numbers on black. When there are no white pixels in the area, it's an inter-digit area, and hence a unit has been consumed, and the used units counter can be incremented.
This isn't useful for instantaneous power consumption, I still use the SDR dongle to show that, but I've found it to be much more accurate - it doesn't work for the transition from 9 to 0 for some reason on my meter - I've not worked out why yet (haven't managed to observe it happen) but I can allow for that by a mod 10 additional increase in the code - and in the few weeks I've been running this, it is mostly spot on, and occasionally out by 1 kWh a week. I am also playing with an OCR version to read the digits just for fun.
May not suit your use case but as I say, it may be another way to "skin a cat"
