The Lab

A personal datacenter running production-grade systems — because security engineers should practice what they preach.

0 Pods
0 K3s Nodes
70TB Storage
0 SIEM Agents
0 MCP Tools

Infrastructure Overview

A full-stack lab: two hypervisors, a DGX Spark for local AI inference, a 70 TB NAS, network segmentation, and monitoring at every layer.

External / Internet
Cloudflare DNS
Tailscale VPN
Let's Encrypt
Edge / Gateway
UDM Pro Max
IPS/IDS
DPI
Traefik Ingress
Compute
Proxmox VE ×2
K3s Cluster (9 nodes)
NVIDIA DGX Spark
LXC Containers
Frigate + Coral EdgeTPU
Storage & Data
Synology DS1821+ (70TB)
NVMe Cache
NFS / SMB
Monitoring & Security
Wazuh SIEM
Prometheus
Grafana
Alertmanager
CrowdSec
OpenBao (HA)

Compute

Two Proxmox hypervisors plus an NVIDIA DGX Spark for local AI inference.

  • 2 Proxmox VE hypervisors
  • AMD Ryzen 9 5950X, ~126 GB RAM
  • AMD Ryzen 9 7950X, ~124 GB RAM
  • NVIDIA DGX Spark (GB10, 121 GB unified memory) serving a local 27B model via vLLM + LiteLLM
  • VMs + LXC containers

Storage

Synology NAS with redundant parity and NVMe caching.

  • Synology DS1821+ (8-bay)
  • 70 TB raw capacity
  • SHR-2 dual parity RAID
  • NVMe read/write cache
  • NFS + SMB for services

Network

Full UniFi SDN stack: gateway, switching, and Wi-Fi.

  • UniFi Dream Machine Pro Max
  • 2 managed PoE switches
  • 6 WiFi access points
  • 8 segmented networks (VLANs)
  • IPS/IDS + DPI enabled

Kubernetes Cluster

A 9-node K3s cluster running ~245 pods with ArgoCD GitOps for fully declarative, version-controlled deployments.

Cluster Topology

High-availability control plane with dedicated worker nodes for different workload types.

Control Plane
ctrl-01
Control Plane
ctrl-02
Control Plane
ctrl-03
Worker
worker-01
Worker
worker-02
Worker
worker-03
Worker
worker-04
Worker
worker-05
Worker
worker-06

GitOps with ArgoCD

Every workload is defined in Git and automatically synced. No manual kubectl applies, no drift.

  • All manifests version-controlled in GitHub
  • Auto-sync with self-healing enabled
  • Application-of-applications pattern
  • Secrets managed via sealed-secrets
Git Push
GitHub
ArgoCD
K3s Cluster
Running

Workload Types

  • Deployments & StatefulSets
  • CronJobs for scheduled tasks
  • DaemonSets for node agents
  • Jobs for one-off operations

Networking

  • Traefik IngressRoute CRDs
  • MetalLB for LoadBalancer IPs
  • CoreDNS for service discovery
  • Network policies per namespace

Storage

  • NFS CSI driver for persistent volumes
  • Local path provisioner
  • Synology NAS backing store
  • Volume snapshots for backups

Network Architecture

Eight segmented networks managed by a UniFi gateway, which also serves internal DNS.

Management Mgmt
Gateway, switches, and access points.
Own network segment Infrastructure devices only
Servers VLAN 30
Proxmox hypervisors, the K3s cluster, the NAS, and platform services.
Wazuh agents on hosts Prometheus monitoring
Family Devices VLAN 40
A separate network for family members' everyday devices.
Own network segment
Trusted VLAN 50
A separate network for trusted personal devices.
Own network segment
IoT VLAN 60
Smart home devices, managed through Home Assistant.
Own network segment Home Assistant integration
Cameras VLAN 70
Isolated camera VLAN: cameras sit on their own segment, recorded by the Frigate NVR.
Isolated camera VLAN Recorded by Frigate NVR
Cluster VLAN 99
Dedicated network for Proxmox cluster (Corosync) traffic.
Cluster traffic only
External / DMZ VLAN 100
Separate firewall zone for externally-facing services; public sites are published through a Cloudflare Tunnel behind the Cloudflare WAF.
Separate firewall zone Cloudflare Tunnel + WAF

Wireless

  • 6 UniFi WiFi 6E access points
  • Band steering & fast roaming
  • Per-SSID VLAN assignment
  • Enterprise WPA3 for management

Switching

  • 2 managed PoE+ switches
  • 802.1Q VLAN trunking
  • STP with root guard
  • IGMP snooping for multicast

DNS & Proxy

  • UniFi gateway (internal DNS)
  • Cloudflare DNS (external)
  • Traefik reverse proxy with auto HTTPS
  • Tailscale VPN (dual HA routers)

Security Stack

Defense in depth: network segmentation, endpoint monitoring with Wazuh, blocking at three layers (UniFi, CrowdSec, Cloudflare edge), and a highly available credential vault.

Wazuh SIEM

Full security information and event management platform with 27 agents across infrastructure hosts. Centralized log analysis, vulnerability scanning, file integrity monitoring, rootcheck, and security configuration assessment.

27 Agents Vuln Scanning FIM Rootcheck SCA Compliance MITRE ATT&CK

Camera Surveillance

Frigate NVR with 7 cameras on an isolated camera VLAN. Object detection runs on a Coral EdgeTPU; an NVIDIA RTX 3080 Ti handles video decode and local vision-LLM descriptions of events (Ollama). Person, vehicle, and animal detection with zone-based alerting and 24/7 recording.

7 Cameras Coral EdgeTPU Vision-LLM Descriptions Zone Alerts 24/7 Recording

Network Security

UniFi IPS/IDS with deep packet inspection at the gateway. Country-based geo-blocking, content filtering, and real-time threat intelligence feeds.

IPS/IDS DPI Geo-blocking Threat Intel

Three-Layer Blocking

Wazuh detections trigger blocks at the UniFi firewall, in CrowdSec, and at the Cloudflare edge. Cloudflare WAF rules are managed automatically across 18 zones, with expiring IP blocks and merge-safe rule updates.

Wazuh CrowdSec Cloudflare WAF 18 Zones

Credential Vault

A 3-node OpenBao cluster (Raft storage, auto-unseal) tolerates the loss of one node and holds the AI platform's credentials, with SSO for human logins and drift detection.

OpenBao 3-node HA Auto-unseal

Certificate Management

Automated TLS certificate provisioning via Let's Encrypt with Traefik. Wildcard certificates for internal services, auto-renewal, and HTTPS everywhere.

Let's Encrypt Wildcard Certs Auto-renewal HTTPS Everywhere

Security Principles

  • Zero trust by default
  • Least privilege access
  • Network segmentation
  • Defense in depth
  • Continuous monitoring
  • Automated remediation
  • Encrypted at rest & in transit
  • Regular vulnerability scans
  • Incident response playbooks
  • Audit logging on everything

Monitoring & Observability

Full observability stack with metrics, alerting, and dashboards for every layer of the infrastructure.

Prometheus

Time-series metrics database scraping every host, service, and application in the stack.

  • Custom scrape configs per service
  • Node Exporter on all hosts
  • Service-specific exporters
  • Long-term retention

Grafana

Visualization platform with custom dashboards for every subsystem in the lab.

  • Infrastructure dashboards
  • Network traffic analytics
  • K3s cluster health
  • Application-specific views

Alerting

PrometheusRule-based alerting with multi-channel notification delivery.

  • CPU, memory, disk alerts
  • Service health checks
  • Network anomaly detection
  • Discord + Telegram notifications

What Gets Measured

  • Host CPU, memory, disk I/O, network
  • K3s pod resource consumption
  • NAS disk health & utilization
  • Camera stream health & FPS
  • Traefik request rates & latency
  • Wazuh alert frequency
  • Cloudflare edge traffic & firewall events
  • UPS battery status

Alert Conditions

  • CPU sustained above 90% for 5 minutes
  • Memory usage above 85%
  • Disk usage above 80%
  • Pod restart count exceeds threshold
  • Service endpoint unreachable
  • SSL certificate expiry within 14 days
  • SIEM critical event detected
  • NAS disk health degradation

Services Catalog

~245 pods running across 9 Kubernetes nodes. Services are deployed via ArgoCD GitOps and monitored by Prometheus.

Media & Entertainment

Plex

Media streaming server with hardware transcoding and remote access

Sonarr

TV series management and automated download orchestration

Radarr

Movie collection management with quality profiles and upgrades

Lidarr

Music library management and automated acquisition

Prowlarr

Unified indexer manager for all media automation services

Overseerr

Request management portal for media library additions

Home Automation

Home Assistant

Central home automation hub with 200+ entities across lighting, climate, and sensors

Node-RED

Visual flow-based automation for complex multi-service workflows

Zigbee2MQTT

Zigbee device bridge supporting 50+ devices without vendor lock-in

WLED

Addressable LED controller for ambient and accent lighting effects

Security & Surveillance

Wazuh SIEM

Security information and event management with 27 monitored agents

Frigate NVR

NVR with object detection on a Coral EdgeTPU; the RTX 3080 Ti handles video decode

CrowdSec

Behavior-based detection and IP bans behind Traefik, alongside the Cloudflare WAF

Monitoring & Observability

Prometheus

Time-series metrics collection with custom scrape configs for every host

Grafana

Visualization dashboards for infrastructure, network, and application metrics

Alertmanager

Alert routing and notification delivery via Discord, Telegram, and email

AI & Automation Platform

Thor MCP Server

Custom AI management platform with 490+ tools across 45 tool groups

Discord Bot

AI-powered bot for infrastructure management via slash commands

Telegram Bot

Mobile-first infrastructure control and monitoring

Ollama

Local vision-LLM on the RTX 3080 Ti that writes descriptions of camera events

vLLM + LiteLLM

Self-hosted 27B model on an NVIDIA DGX Spark, the last-resort provider in Thor's LLM failover chain

Documents & Productivity

Paperless-NGX

Document management system with OCR, tagging, and full-text search

Vikunja

Task management and project tracking with Kanban boards

Infrastructure & Networking

Traefik

Reverse proxy and ingress controller with automatic Let's Encrypt certificates

ArgoCD

GitOps continuous deployment for all Kubernetes workloads

Cloudflare DNS

External DNS management with DDoS protection and CDN

Tailscale VPN

Zero-config mesh VPN with dual subnet routers for high availability

AI Management Platform

A custom-built AI platform (Thor) that manages the lab through 490+ MCP tools, 13 specialized agents plus 2 review roles, and multi-channel interfaces.

Thor MCP Server

Model Context Protocol server providing AI assistants with direct access to every system in the infrastructure. Built on Claude AI with Anthropic's SDK.

  • 490+ tools across 45 tool groups
  • Proxmox, K3s, NAS, network, DNS control
  • Security scanning & SIEM integration
  • Media stack management
  • Home automation control
  • Monitoring & alerting management

Agent Architecture

13 specialized AI agents plus 2 review roles, each with scoped tool access and domain expertise. Hot-reloadable registry with persistent memory.

  • Thor (Overseer) — primary coordinator
  • Heimdall — security monitoring
  • Jarvis — infrastructure management
  • Ada — home automation
  • Apollo — media stack
  • Offsec — pentesting & red team
  • + 7 more specialized agents

Interfaces

  • Claude Code CLI (primary)
  • Discord bot with slash commands
  • Telegram bot for mobile
  • Web dashboard (HTMX + Hono)

Memory System

  • Vector embeddings for semantic search
  • Daily logs & session tracking
  • Topic-based knowledge base
  • Incident history & resolution

AI Hardware

  • NVIDIA DGX Spark (GB10, 121 GB unified)
  • Local 27B model via vLLM + LiteLLM
  • RTX 3080 Ti: video decode + Ollama vision-LLM
  • Coral EdgeTPU for camera object detection

Platform Engineering Highlights

  • 3-node OpenBao HA credential vault (Raft, auto-unseal)
  • Cloudflare WAF automation across 18 zones: Wazuh detections trigger blocks at UniFi, CrowdSec and the Cloudflare edge
  • Tiered auto-remediation engine: pattern library, LLM risk judge, tiers A/B/C, circuit breakers, and post-fix verification
  • LLM failover chain Claude → Codex → self-hosted Qwen on the DGX Spark, with active health probes and breakers
  • Observability by default: every service ships Prometheus metrics, alert rules, and a Grafana dashboard

Maker Space

Beyond software: a workshop for physical prototyping, custom fabrication, and creative projects.

Laser Cutting

xTool P1

40W diode laser for cutting and engraving wood, acrylic, leather, and more. Enclosed workspace with exhaust filtration.

Fiber & Diode Laser

xTool F1 Ultra

Dual-laser system (fiber + diode) for engraving metals, plastics, wood, and stone. High-speed galvo marking.

FDM 3D Printing

Bambu Lab X1-Carbon

Multi-material FDM printer with AMS (Automatic Material System). Carbon fiber, PETG, TPU, and 16-color prints.

Resin 3D Printing

Elegoo Saturn 16K

16K resolution MSLA resin printer for ultra-detailed miniatures, prototypes, and jewelry molds.

UV Printing

eufyMake E1

Desktop UV printer for full-color, raised-texture prints directly onto wood, acrylic, metal, leather, and other flat or curved surfaces.

Vinyl Cutting

Cricut

Precision cutting machine for vinyl, paper, fabric, and thin materials. Custom stickers, decals, and craft projects.

By the Numbers

The scale of a personal infrastructure, measured.

0
Pods
Across 9 K3s nodes
0+
MCP Tools
Across 45 tool groups
0TB
Storage
SHR-2 protected NAS
0GB
RAM
Across two Proxmox hypervisors
0
SIEM Agents
Wazuh-monitored endpoints
0
Cameras
Coral EdgeTPU object detection
0
Networks
Segmented VLANs
0
AI Agents
Plus 2 review roles
0
CPU Threads
Ryzen 9 5950X + 7950X
0
Maker Machines
3D printers, lasers, UV printer, vinyl cutter

Built to Learn, Secured to Last

This homelab is a living platform for continuous learning in security, infrastructure, and automation.

Get in Touch Back to Resume