Doremon: Controlling My Home With Brain Signals

I built Doremon at a hackathon: an EEG-based home automation system that lets you toggle appliances and multimedia without touching anything — just by looking at a flickering light. It was equal parts hardware hacking and signal processing, and it taught me a lot about how noisy and unforgiving real brain signals are.

The goal wasn't a polished product. It was to see whether I could take raw EEG, extract intent from it, and act on that intent within the constraints of a hackathon weekend.

Why I made Doremon

Brain-computer interfaces sound like science fiction, but the underlying idea is surprisingly approachable if you pick the right signal. I wanted a project that forced me to touch the whole stack:

• capture real EEG from a person,
• turn that raw signal into a decision,
• and drive physical hardware from that decision.

Doremon was my way of collapsing all three into one loop that actually did something in the physical world.

How it works: SSVEP

The trick that made this feasible is SSVEP — Steady-State Visually Evoked Potentials.

When you stare at a light flickering at a fixed frequency, your visual cortex produces electrical activity at that same frequency. So if I put two lights in front of you, one flickering at one rate and one at another, I can look at your EEG, find which frequency dominates, and infer which light you're looking at.

That's the whole intent-detection mechanism:

• Each appliance maps to a flicker frequency.
• You look at the light for the thing you want to toggle.
• The pipeline reads the EEG, finds the matching frequency, and fires the action.

It's a clever shortcut — instead of decoding complex thoughts, I only had to detect where you're looking from the frequency signature in the signal.

What I built and what I learned

I built both halves myself:

• The hardware — the flickering visual stimuli and the wiring to actually switch appliances and multimedia on and off.
• The signal-processing pipeline — taking the raw EEG and classifying which fixed frequency the person was attending to.

The honest result: about 67% intent accuracy. That's well above chance, enough to demo convincingly, but nowhere near something you'd trust to control your actual house.

A few things stuck with me:

1. EEG is loud and messy. A huge fraction of the work is fighting noise, not decoding intent. Muscle movement, blinks, and electrical interference all bury the signal you care about.
2. Picking the right paradigm matters more than clever code. SSVEP works because it turns a hard decoding problem into a much easier frequency-detection problem. The choice of signal did more for accuracy than any algorithm I wrote.
3. 67% is a real number, not a failure. It's a reminder that BCIs are probabilistic. Even good systems get it wrong often enough that the interface has to be designed around uncertainty.

Doremon didn't become a product, and I'm not going to pretend the accuracy was production-ready. But it made brain-computer interfaces concrete for me: a signal I could see, a frequency I could detect, and a light bulb that turned on because of it.

If you want to look at the code, here it is: Doremon on GitHub.