128GB ram, 2x rtx6000 blackwell, lots and lots of storage, but stuck in paralysis choosing a harness / model

I have been accumulating parts for over 8 months to merge all of my existing hardware into one massive workstation. I finally had the time to do it, and while I use the GPUs for algo trading from 9:30 to 4:15 p.m., I want to start to use local AI from 5pm on. I'm at a point where I could see myself having a really fulfilling hobby reverse engineering, firmware hacking, and just experimenting with building Frankenstein gadgets from my tech graveyard. There are many experiments I would like to perform on financial data as well that would just take too long for me to manually code and I am not interested in having these thoughts and experiments be in a large provider's training set. I've been lurking on this sub for a long time and have my own little agent I use for some specific tasks using gemma3 but that was on my 3090 in C# + llama.cpp. So my long-winded question, what thick desktop client will give me the most seamless way to build autonomous agents, play with new models, and what version of qwen 3.8 flash should I be using with my hardware, assuming that is the right model.