This project is a homelab edge datacenter with k3s deployed, hosting a voice-enabled 3D assistant and other open-source components. It emphasizes recycled hardware, self-hosted sovereignty, and cost-efficient infrastructure for real-time AI and monitoring workloads.

This Mini-Datacenter is composed of:

• 4× Intel NUCs (reconditioned)
• 1× Raspberry Pi 5 + AI Hat Kit (Optimized for CNN inference acceleration)

• 1× Reconditioned QNAP NAS with:
  • 6× HDD drives
  • MinIO as S3-compatible storage backend

• TP-Link Smart Managed Switch (VLANs and isolation)

• SnortML IDS (experimental setup for Zero-Day threat detection)
• Prometheus + Grafana stack for real-time observability
• Dashboard on recycled iPad (Grafana front)

flowchart TD
  subgraph EdgeCluster["🏠 Home Datacenter Edge Setup"]
    NUC1(NUC 1)
    NUC2(NUC 2)
    NUC3(NUC 3)
    NUC4(NUC 4)
    PI5(Raspberry Pi 5 + AI Hat)
    NAS(QNAP NAS + MinIO)
    Switch(TP-Link Managed Switch)
    IDS(SnortML IDS)
    Grafana[iPad Display with Grafana]
  end

  NUC1 -->|K3s Node| Switch
  NUC2 -->|K3s Node| Switch
  NUC3 -->|K3s Node| Switch
  NUC4 -->|K3s Node| Switch
  PI5 -->|Audio & AI Tasks| Switch
  Switch --> NAS
  IDS --> Switch
  NAS -->|S3 API| NUC1
  NAS -->|S3 API| NUC2
  NAS -->|S3 API| NUC3
  NAS -->|S3 API| NUC4
  Switch --> Grafana

.
├── examples
│   ├── transcription
│   │   ├── whisper.py
│   │   └── youtube_wav_fetcher.py
├── manifests
│   ├── minio
│   ├── monitoring
│   ├── ollama
│   └── whisper
├── src
│   ├── backend
│   │   └── watchservice
│   │       └── main.py
│   ├── frontend
│   │   ├── components
│   │   │   ├── Avatar.jsx
│   │   │   └── Experience.jsx
│   │   ├── App.jsx
│   │   └── main.jsx
├── text-to-speech-webgpu
│   ├── src
│   │   ├── App.jsx
│   │   └── main.jsx
│   └── index.html
├── package.json
└── README.md

curl http://localhost:11434/api/pull -d '{
  "name": "smollm:135m"
}'

curl http://localhost:11434/api/generate -d '{
  "model": "smollm:135m",
  "prompt": "What is your name"
}'

https://dan-niles.medium.com/setting-up-apache-solr-on-kubernetes-with-rancher-desktop-931433d8f56b

The 3D assistant is built using three.js and react-three-fiber, featuring an animated avatar that lip-syncs and responds to user inputs.

• Avatar.jsx: Manages 3D avatar animations and lip sync.
• Experience.jsx: Orchestrates the scene, lights, and environment.

cd src/frontend
npm install
npm run dev

Openvoice OS provides a self-hosted voice assistant backend for speech recognition and responses.