All work
AI Lab Ongoing

Building with AI: From Enterprise to Home Inference

A personal AI lab that bridges enterprise AI experience with hands-on infrastructure. Running production-class models locally to build, learn, and ship real projects.

Hardware setup

GPU

2× RTX 3090

48 GB VRAM combined

Model size

70B params

Llama 3.1 family

Hypervisor

ProxMox VE

Local virtualization

Projects

Foundation

VPN Tunnel to Oracle Cloud

[Placeholder: Overview of setting up a persistent VPN tunnel between home infrastructure and Oracle Cloud free tier. Why this matters for secure access, hybrid compute, and cost-efficient GPU bursting.]

pfSense WireGuard Oracle Cloud ProxMox Vast.AI

Secure hybrid network linking on-prem and cloud GPU resources at zero recurring cost.

Infrastructure

Self-hosting Llama 3.1 70B

[Placeholder: How to run a 70B parameter model locally on dual RTX 3090s with vLLM — quantization choices, VRAM management, throughput benchmarks, and what workloads this unlocks without cloud API costs.]

Llama 3.1 RTX 3090 CUDA vLLM

Full 70B inference at home — zero API costs, full data privacy, usable throughput for real tasks.

Application

AI Video Upscaling with Claude Code

[Placeholder: Using Claude Code to build a Real-ESRGAN + FFmpeg pipeline for upscaling archival video footage. The workflow, the prompting strategy, the quality tradeoffs, and the results on actual source material.]

Before

Original footage

After

AI upscaled to 4K

Claude Code Python Real-ESRGAN FFmpeg

Automated pipeline upscaling archival footage to 4K — built with Claude Code in a single session.