Built my 10x NVidia V100 AI Server - 320gb vram - vLLM Testing Linux Headless - Just a Lawyer,Need Tips
South Carolina attorney spent months building private AI server with 12 V100 GPUs for local legal automation.
A South Carolina trial lawyer has documented his four-month journey building a custom AI server from scratch, despite having no prior experience with hardware assembly or Linux administration. His system features 10 NVIDIA V100 SXM2 GPUs with 32GB VRAM each (320GB total), connected via NVLink bridges on a Threadripper Pro platform with 256GB DDR4 RAM. The lawyer plans to expand to 12 V100s for 384GB total VRAM, creating what he calls his "corniest mid-life crisis" but also a serious tool for legal automation.
Using Claude Code (Claude 3 Opus) to orchestrate the entire setup process, he's been testing vLLM inference engines with various local models to automate paralegal tasks like document analysis, form filling, and pattern recognition. His goal is to find models that can emulate his writing style and perform legal reasoning, potentially using QLoRA/DORA fine-tuning techniques. The project highlights how professionals outside traditional tech fields are now building custom AI infrastructure to gain privacy, control, and domain-specific capabilities that cloud services can't provide.
- Built 10x NVIDIA V100 server with 320GB VRAM (expanding to 384GB) using Threadripper Pro platform
- Used Claude Code (Claude 3 Opus) to orchestrate entire Linux/vLLM setup despite no prior engineering experience
- Aiming to automate legal tasks like document analysis and form filling with local QLoRA-tuned models
Why It Matters
Shows professionals can build enterprise-grade AI infrastructure without engineering backgrounds, enabling private domain-specific automation.