Smart Home Lab: Complexity as a Research Platform
The Problem
Smart home devices are easy to buy and hard to orchestrate. Every manufacturer has an app. Every app has its own automation model. The result is a house full of “smart” devices that don’t know about each other and a phone full of apps that each control one thing.
Home Assistant fixes the integration problem. It talks to everything. But at scale - real scale, hundreds of entities across dozens of actual devices - the automation layer becomes its own engineering challenge. How do you write automations that are reliable, debuggable, and don’t cascade into chaos when one sensor reports a bad value?
What We Built
A complex and frankly unwieldy Home Assistant installation that serves as equal parts home automation system and research platform. The public repository contains YAML configurations and Python scripts that handle the non-obvious parts: the patterns, templates, and architectural decisions that emerge when you have more sensors than common sense.
The dashboard is hand-built - not auto-generated. It tracks solar production, battery state of charge, grid draw, self-sufficiency percentage, per-room temperatures, AC controls, weather forecasts, and energy cost in real time. Sub-dashboards break out different concerns: energy, climate, devices, and an AI-focused view for the experimental stuff.
Where the Projects Converge
This is where several Throughline projects stop being abstractions and start touching hardware.
The Johnny Solarseed rate calculator generates time-of-use rate configurations - off-peak, mid-peak, on-peak pricing with all the regulatory surcharges baked in - and exports them as YAML that plugs directly into Home Assistant. The peak shaver algorithm uses those rates, tomorrow’s solar forecast, and current battery state to decide when to charge from the grid, when to hold, and when to discharge. The goal: minimize cost by shifting load to cheap hours and maximizing self-sufficiency during expensive ones.
The “Solarseed Peak Shaver Recalculate” button on the dashboard triggers a re-evaluation whenever conditions change - a cloudy forecast replaces a sunny one, or the battery didn’t charge as expected. It’s the feedback loop between prediction and reality.
AI as a Co-Pilot
We’re using AI to help wrangle the complexity in two ways. First, a dedicated AI dashboard in Home Assistant that helps reason about system state - surfacing patterns and anomalies across the sensor data that would take forever to spot manually. Second, AI-assisted development of the peak shaver algorithm itself. The optimization problem (when to charge, hold, or discharge given forecasted solar, weather, TOU rates, and battery constraints) is exactly the kind of multi-variable decision-making where AI collaboration accelerates the iteration cycle.
The Philosophy
Every automation should be deletable. If removing an automation makes the house broken rather than just less convenient, the automation is doing too much. The house should work without software. The software should make it work better.
This constraint prevents the worst failure mode in home automation: a system so complex that a server reboot means you can’t turn on the lights. The lights work. The automations just make them smarter about when they turn themselves on.
The Throughline
The Smart Home Lab is where ideas stop being abstractions and start touching hardware. Johnny Solarseed’s rate calculator feeds real data into Home Assistant. The peak shaver algorithm makes real decisions about real batteries. The Process captures what we learn. Every project on this site has a wire running through this lab somewhere.
Current Status
Active and perpetually evolving. The solar optimization loop is the current focus - getting the peak shaver algorithm to reliably beat a naive “charge when the sun’s out” strategy. The installation grows steadily as new devices and platforms appear, and the repository is updated when patterns stabilize enough to be useful to others.