This is the physical infrastructure I run at home. High-throughput AI workstations, virtualization nodes, and edge routers — every deployment, model inference, and stack runs on these machines. No cloud dependency unless required.
Why this matters for your business: This lab demonstrates real-world capabilities in local GPU acceleration, high-throughput model inference (vLLM), DNS management, reverse proxying, system hardening, and strict uptime maintenance without third-party dependencies.
I believe in maintainable infrastructure. See how some of these systems are deployed and documented in my technical blog, demonstrating clear knowledge transfer for when I hand over a project to your team.
Workstation & AI Node
MSI Crosshair A16 HX (AMD Ryzen 9)
Online / Active
CPU
AMD Ryzen 9 8940HX
16 Cores / 32 Threads @ up to 5.39 GHz
GPU Acceleration
NVIDIA RTX 5070 Laptop
8 GB GDDR7 VRAM — Tensor & CUDA Cores
System RAM
64 GB DDR5
Heavy workloads, KV Cache offload & compilation
OS & Environment
Fedora Linux + Hyprland
Pure Wayland, minimal idle resource footprint
Storage
Primary NVMe
1 TB
NVMe SSD (Model Weights & OS)
Secondary Drive
512 GB
NVMe SSD (Gaming: Rainbow Six Siege & Autodesk Fusion — no Linux support)
Host Purpose & Workload: Personal primary daily machine (MSI Crosshair A16 HX laptop) and dedicated local AI workstation. Used for model training, fine-tuning, quantization benchmarking (AWQ / GGUF), and local low-latency inference testing without cloud overhead. Includes a secondary 512 GB drive dedicated strictly to gaming (Rainbow Six Siege) and Autodesk Fusion CAD (unsupported on Linux). Runs Syncthing to stream automated backups to the HP EliteDesk server over Tailscale mesh network.
Performance Node
HP EliteDesk 700 G4
Online
CPU
AMD PRO A10-8770E R7
10 Compute Cores — 4C + 6G
RAM
16 GB
OS
Debian 13 (Trixie)
Primary server
Storage
System
128 GB
SSD
Multimedia 1
1 TB
NVMe SSD
Multimedia 2
512 GB
NVMe SSD
Backup & Data
4 TB
HDD
Host Purpose & Status: Performance-oriented node for heavier homelab workloads (ERP, Docker dev environments, backup target with Syncthing over Tailscale from the MSI PC, and media streaming). It delivers compute power for local services, though it occasionally becomes unresponsive under certain loads or hardware states.
High Availability Node (24/7)
Raspberry Pi 4B
Online / Always On
CPU
Broadcom BCM2711
Quad-core ARM Cortex-A72 @ 1.8 GHz
RAM
4 GB LPDDR4
OS
Debian 13 (Trixie)
Lightweight services & hardware integration
Host Purpose: High availability node engineered to stay always on (24/7). Runs critical network and security infrastructure (Cloudflare tunnels, NPM reverse proxy, AdGuard Home DNS filtering, Gitea, and monitoring) ensuring uninterrupted reliability.
MSI Crosshair A16 HX (AMD Ryzen 9)
AI & Local Compute
vLLM InferenceHigh-throughput GPU inference engine with PagedAttention and metricsCUDA / Local
Ollama & Open WebUILocal developer model serving & rapid prompt testing environmentInternal
Model QuantizationAWQ & GGUF quantization pipelines for optimal VRAM offloadingToolkit
Hyprland DesktopMinimalist Wayland shell tuned for low idle RAM and max available VRAMFedora Host
SyncthingContinuous backup synchronization to HP EliteDesk server over Tailscale meshTailscale / P2P
HP EliteDesk 700 G4
Business & Infrastructure
SyncthingBackup receiver and file synchronization daemon from MSI workstation via TailscaleTailscale / Private
This lab is where I validate architectures before bringing them to a client's production environment. If your company needs a reliable, well-documented setup like this: