When I left Elastic, one of my goals was to dive deep into home automation and finally “level up” my setup. Looking back at where I started when we first moved to Graham, it’s almost comical.
Back then, “Home Automation” was just a few Philips Hue bulbs and a Raspberry Pi running Home Assistant that didn’t do much. It was essentially a glorified light switch connected to Alexa.
Fast forward to today, and the Raspberry Pi is a distant memory, replaced by a stack of enterprise-grade hardware and local AI. Here is the state of the union for my home lab as we kick off the new year.
The Backbone: Networking & Infrastructure
You can’t build a smart home on a shaky foundation. I’ve moved away from consumer “plug-and-play” and into a robust, segmented network that handles everything from 10G fiber runs to rural failover.
- A Synology RT6600ax handling a dual-WAN setup. We use Xfinity for the heavy lifting and Verizon Fios for failover because “offline” isn’t an option when you both work remotely. Synology mesh nodes extend WiFi for a unified network across the property.
- Cisco managed and unmanaged switches. I’ve got 10G links running between the house and the shop as well as the primary NAS and 2.5G to the servers, ensuring that moving large files or camera feeds never hits a bottleneck. Everything is tucked away in its own VLAN. My IoT devices can’t talk to my primary machines, and my security cameras live in their own private sandbox.
- Synology NAS devices act as my primary network storage and what I call my “on-site-off-site” backup. The small NAS in the shop gets a nightly backup of core files from the main rackmount NAS under the stair.
The VLAN Layout
I still use 1 as the primary for devices like computers, phones and laptops, but I have gotten more intentional about using VLANs to keep the noise down with over 100 devices connected.
| IoT (30) | Smart plugs, light bulbs, and things I don’t fully trust. |
| Security (40) | All camera traffic and NVR feeds. |
| Work (60) | My professional life, isolated from the rest of the chaos. |
Home Assistant: The Central Nervous System
I didn’t go straight from a Pi to a server. There was a brief “middle age” where I ran everything in Docker on my Synology NAS. It worked, but eventually I got so comfortable with CLI docker that I got frustrated with using it in Synology DSM.
Today, Home Assistant lives on a dedicated server managed by Portainer. It is the rock-solid core of our daily lives. We’ve gone from “it turns on lights” to over 40 complex automations that manage:
- The HVAC in the house and shop, heaters and fans
- Lighting using motion and doors all over
- The chicken coop (yes, the chickens have their own tech stack)
- Geofencing for locks and garage doors
- Notifications from Frigate
Frigate: The Eyes of the Property
I’ve always been frustrated with “AI” cameras that think a swaying tree branch is a burglar. While my Synology Surveillance Station handles the 24/7 recording, Frigate is the brains.
Using a Google Coral TPU for inference and an RTX 3050 for encoding, it performs local object detection on 21 different camera feeds without breaking a sweat. When someone drives onto the property, I get a rich notification on my watch within a second. No cloud, no subscriptions, no data leaving the house.
The “Jarvis” Phase: Local AI & LLMs
This is where things get really fun. I’ve built a dedicated AI server using a 12-core i7 and dual GPUs (RTX 4070 & 2080ti). Using n8n, Ollama and Fast API, I’m running a private intelligence layer over the whole house.
- Nova: My shop assistant. It’s not quite Jarvis yet, but it can control the house, take notes in Obsidian, and answer questions about my project history using RAG.
- The Second Opinion: When Frigate detects an object, my AI server “double-checks” the frame with a higher-quality model to reduce false positives. This is much more effective at reducing noise than the complex per-camera configurations in Frigate.
- The Recipe Agent: I scan a printed recipe; the AI OCRs it, formats it, and injects it into Mealie.
The Full Stack (Everything Else)
To keep all of this moving, I use a additional suite of self-hosted and cloud services. These are split across the two servers I mentioned, and a third HP mini desktop PC. It is running ubuntu server and hosts docker containers. It runs Portainer to manage docker across all three servers.
- Grafana/InfluxDB: Home Assistant sensor history
- SearXng: Search for AI agents
- Plex: Home Media Server
- Stirling PDF: Document Handling
- Qdrant: Vector Database for RAG
- Ollama/Open WebUI: LLM Tools
- n8n: Cloud Workflows with Cloudflare to local Ollama via Open WebUI Proxy
- Gitea: Source control, docker compose, frigate and home assistant configuration
- Immich: Phone photo backups
- Mealie: Recipes
- PicoShare: Local image hosting
- TwinGate: External access to Frigate and Home Assistant
- Cloudflare: External Access to Open WebUI & Ollama
- Portainer: Container management for all servers
- Beszel: Performance monitoring for all servers
- Homepage: Home Dashboard
- Obsidian Sync: Keeping my notes synced for vector encoding
Final Thoughts
Looking back, the biggest change is the utility. Home automation used to be a hobby I spent time fixing. Now, it’s an invisible utility that saves us time and provides security. Whether it’s the chickens being tucked in automatically, the doors locking when we leave or the AI summarizing my day’s notes, the “Smart Home” is finally starting to feel smart.
What’s next for 2026? Getting my cloud and local worlds better integrated. I started working with n8n locally, and quickly fell into OAuth issues with services like Google Docs and Trello. I moved to n8n cloud, with Cloudflare now setup to access Open WebUI on the public internet, my next step is to complete the flow so I can use n8n cloud for orchestration and my local AI server for LLMs to take advantage of my own hardware.
