It seems that an important part of the strength of engines like AI or Zeus is related to their specific nnue nets(Zeus nets). As mehmet mentioned before providing neural networks needs to have very sophisticated hardwares. So it’s very difficult to compete with these kinds of engines. Like Zeus,AI,and fat titz. And needs to work very hard on project. And may be need a complete teamwork. I wish you success Alex.
Finally I have found a way to test engines in a more scientific way, using DOTS4 unbalanced pgn 400 positions opening test suite and TCEC 7 balanced 306 positions (2014, so after 8 years consumer hardware probably matches TOP PC speed requisites) switching sides for each position. I will test Proteus Chess SF v006 against Fat Titz 2 both popcnt sse41 con Windows 11 ARM64 with these opening books. Now Polyglot is almost useless for me, since BSG let switching sides only using PGN and .bin often suddenly stops loading on BSG after crashes.
Right now i prefer handcrafted fine tuning Proteus instead using Fishtest or other tools, but I'm studying them. I'm also adding learning to next Proteus releases.