Nice, I did a similar thing with OCP documents and an SXM2 server schematic set I found, but ultimately lost interest in the project after I rented a server with 8x V100s and realized it was kind of crap for LLM inference.