Software project · smart home · 2026
Home assistant
Talk to your home like a person. Safe actions run at once, critical ones ask for confirmation, wrong ones are stopped.
live demo Simulation with sample rooms
Rooms and devices in the demo are made up: no connection to a real home.
- One interfaceDashboard, remote and voice assistant act on the same devices, with state updated in real time.
- The AI proposes, the rules decideA deterministic validator separates safe actions from critical ones and blocks out-of-range values.
- Credentials stay safeThe Home Assistant token stays on the server: the browser never sees it.
A — The problem
Home assistants are convenient until they get it wrong. Opening the gate or starting a pump because of a misheard command is not a small mistake.
B — What I built
An assistant connected to Home Assistant, with a web dashboard, a remote and a voice chat with a wake word. Every command — from a button or from the AI — goes through a deterministic validator: an allowlist of actions, limits such as temperatures between 16 and 28 °C, and explicit confirmation for gates, irrigation, pump and water heater. The home’s credentials never leave the server.
C — How it works
- 01
The command
From a button, chat or voice.
- 02
The validator
Allowed, needs confirmation or denied: code decides, not the model.
- 03
Confirmation
Critical actions open a sheet with the reason and two buttons.
- 04
Real state
The dashboard updates live from the home’s events.
D — The system
A simplified map of the components: not the technical documentation, but it shows how the parts talk to each other.
E — Why it works
Safety outside the model
Rules live in code and apply to everyone.
Same rule for AI and buttons
There is no shortcut that skips confirmation.
Real-time state
WebSocket from the home to the interface.
Voice at home
Wake word in the browser and spoken answers.