Open Source Home Automation & AI Infrastructure

Field: IT Infrastructure / Home Automation / Artificial Intelligence

Period: 2025 (current deployment)

Status: In production — daily personal use

Location: Playa del Carmen, Mexico

Background

In 2025, I designed and deployed a complete, 100% open source home automation and AI infrastructure for my home in Mexico. The goal: total autonomy, zero cloud dependency, absolute privacy, and intelligent control of the entire home environment.

This stack also serves as an R&D lab for my commercial developments (the professional home automation packs I offer to clients).

Problem

Technical challenges:

• Build a local home automation infrastructure (no Google/Amazon/Apple cloud)
• Integrate conversational AI for smart voice control
• Manage 30+ IoT devices (lights, plugs, air conditioners, sensors, front door)
• Electrical monitoring: current sensors in the electrical panel to analyze consumption
• Network isolation: separate IoT, home LAN and servers
• Performance: voice recognition <2s, real-time automations
• Reliability: infrastructure running 24/7 with no intervention

Privacy challenges:

• Zero personal data in the cloud (no Google/Amazon surveillance)
• Local voice recognition (no external servers)
• Home automation data stays at home

Solution

Infrastructure Architecture

Hardware:

• Main server: 1 physical PC running Ubuntu 24.04 LTS (12-core AMD Ryzen)
• Network: BPI-R4 OpenWrt router with VLANs (isolated IoT)
• Electrical monitoring: current sensors in the electrical panel
• Bluetooth Proxy: ESP32 (Squeezbox server for multi-room audio)
• Storage: 1TB NVMe

Software stack (all open source):

1. Home Assistant OS (KVM VM)

• Complete home automation orchestration
• 30+ devices integrated (Tuya/Zigbee/Z-Wave)

• Controlled equipment:

  • Air conditioners (4 units)
  • Lights (15+ points)
  • Front door (smart lock)
  • Smart plugs
• Energy monitoring: current sensors in the electrical panel (real-time consumption analysis)
• Complex automations (multi-level conditions, multiple triggers)
• Unified web/mobile dashboard
• Local voice recognition: Whisper (STT) + Piper (TTS)
• Offline operation: automations keep running without internet

2. AdGuard Home (Docker)

• DNS blocking: ads, trackers and malware blocked network-wide
• Privacy: no external tracking
• Performance: fast local DNS resolution
• Automatic filtering of 100,000+ malicious domains

3. AnythingLLM (Docker)

• Conversational AI interface
• RAG (Retrieval-Augmented Generation) on local documents
• OpenAI API integration for complex reasoning
• Contextual memory: the assistant remembers our conversations

4. ESP32 Squeezbox Server (BT Proxy)

• Multi-room audio server
• Bluetooth proxy to integrate BT devices into Home Assistant
• Synchronized audio streaming

5. Isolated network (OpenWrt BPI-R4)

• 3 VLANs: home LAN (192.168.1.0/24), IoT (192.168.10.0/24), Servers
• WireGuard VPN: secure remote access
• QoS: traffic prioritization (streaming never stutters)
• Smart firewall: IoT fully isolated from the rest of the network

Technical Workflow

Smart voice recognition

Example interaction:

Me: "It's hot"

Assistant:

• Turns on the living room AC (24°C)
• Closes the front door (if open)
• Turns off unnecessary lights (energy saving)
• Starts a "chill" playlist on Squeezbox

Tech stack:

1. Whisper (local STT): captures my voice → text (~800ms)
2. ChatGPT API (reasoning): analyzes intent + context (~1-2s)
3. Home Assistant: runs the automations (AC, door, lights, audio)
4. Piper (local TTS): replies by voice (~400ms)

Total: ~2-3 seconds from voice command to execution.

Advanced home automations

Real-world examples:

Scenario 1: Smart electrical monitoring

• Current sensors in the electrical panel

• IF consumption >80% of panel capacity THEN:

  • Mobile notification "High electrical load"
  • Automatically cuts non-priority devices
  • Historical consumption log per circuit

Scenario 2: Adaptive thermal comfort

• IF indoor temperature >28°C AND nobody home THEN:

  • Turns on AC in eco mode (26°C)
  • Turns off lights automatically

• IF return home detected (front door opened) THEN:

  • AC switches to comfort mode (24°C)
  • Turns on entrance lights

Scenario 3: Security & savings

• IF front door open >5 min AND person detected outside THEN:

  • Mobile notification "Door open unusually long"
  • Turns off AC (avoids waste)

• IF door closed THEN:

  • Restores AC to its last mode

PCIe Passthrough — Dedicated VM NIC

Technical challenge: Home Assistant needs direct access to the IoT network (VLAN 10) without going through the Ubuntu host.

Solution: PCIe passthrough of a physical network card (NIC2) directly into the Home Assistant VM.

Implementation:

• Enabling IOMMU (kernel)
• Binding NIC2 to the VFIO-PCI driver
• KVM VM with NIC2 in direct passthrough
• Result: the VM has direct hardware access = minimal network latency + perfect isolation

Results

Deployed infrastructure:

• Ubuntu 24.04 LTS server (12-core AMD)
• Home Assistant OS in a KVM VM (4GB RAM, 4 vCPUs)
• AdGuard Home + AnythingLLM in Docker
• ESP32 Squeezbox Server (BT Proxy + Audio)
• OpenWrt router with VLANs + VPN
• 30+ IoT devices controlled
• Real-time electrical monitoring (current sensors in the panel)

Active features:

• Local voice recognition <2s
• Air conditioning control (4 units)
• Lighting control (15+ points)
• Front door control (smart lock)
• Electrical consumption monitoring per circuit
• Multi-room audio (Squeezbox)
• Complex multi-condition automations
• Network-wide ad/tracker blocking
• Secure remote access (WireGuard VPN)
• Unified home automation dashboard
• Conversational artificial intelligence

Personal impact:

• Total autonomy: zero dependency on Google/Amazon/Apple
• Absolute privacy: my data stays at home
• Everyday comfort: a home that anticipates my needs
• Savings: energy optimization (consumption monitoring + smart automations)
• Reliability: works offline, never a cloud outage

R&D benefit:

• Living lab: I test before offering it to clients
• Field validation: the stack runs 24/7 with no intervention
• Commercial foundation: reusable architecture for client packs (Pack 1, 2A, 2B)

What this project demonstrates

• Ability to design a complete IT infrastructure (network, servers, virtualization, security)
• Mastery of advanced open source technologies (KVM, PCIe passthrough, Docker, OpenWrt)
• Systems approach: integrated network + home automation + AI + energy monitoring + security
• Privacy-first: everything local, no cloud dependency
• Applied R&D: a personal lab that becomes a commercial offering

Complete Tech Stack

OS & Virtualization:

• Ubuntu Server 24.04 LTS (host)
• KVM/QEMU (hypervisor)
• Home Assistant OS (VM)

Home automation:

• Home Assistant Core 2024+
• Integrations: Tuya, Zigbee, Z-Wave
• Controls: air conditioners, lights, front door, plugs
• Monitoring: electrical current sensors (panel)
• Automations: advanced conditional logic

AI & Voice:

• Whisper (local STT, OpenAI model)
• Piper (local TTS)
• ChatGPT API (external reasoning)
• AnythingLLM (local RAG)

Audio & Bluetooth:

• ESP32 Squeezbox Server
• Bluetooth Proxy (BT device integration)
• Multi-room audio

Network & Security:

• OpenWrt (BPI-R4 router)
• VLANs (3 isolated zones)
• AdGuard Home (DNS blocking)
• WireGuard VPN
• Firewall + QoS

Containers:

• Docker Engine + Docker Compose
• AdGuard Home container
• AnythingLLM container

Hardware:

• 12-core AMD Ryzen
• 64GB RAM
• 1TB NVMe
• 2x Gigabit NICs (1 host, 1 passthrough)
• Current sensors (electrical panel)
• ESP32 (BT Proxy + Squeezbox)

Previous

Next

Photos/Visuals

Tags: lab, home automation, home assistant, open source, privacy, AI, infrastructure, KVM, docker, OpenWrt, ESP32, energy monitoring, Mexico, 2025