My own AI. In a helmet. On purpose.
Jarvis didn't stand a chance.
There's a scene in the first Iron Man where Tony Stark walks into his garage, says "JARVIS, run a level one diagnostic," and his entire world comes to life around him. The system knows who he is, what he's building, what he needs before he asks.
Every time I watch it, I feel the same thing. Not wonder. Not amazement. Envy.
I've spent two years telling executives, operators, and founders that AI is not some distant future — it is infrastructure. That the question is not whether to build, but how fast, how deep, how deliberately. That the people who figure this out in the next eighteen months will look, in ten years, like the people who figured out the internet in 1997.
I believed every word. I was also going home every night and asking Siri what the weather was.
This needed to stop.
<img src="your-photo.jpg" alt="..." />
Here is what I thought was going to happen: I would buy a small computer board, connect it to my AI system, write a quick piece of firmware, and have a working voice assistant by the weekend. Maybe Saturday. Definitely by Sunday.
What actually happened was I spent forty-five minutes trying to figure out why a cable that looked exactly like a data cable was not, in fact, a data cable.
It was a charging cable. An imposter. It had been sitting in my cable drawer for what I can only assume was years, presenting itself as useful, contributing nothing, living a lie.
This is called a "charge-only cable." It is genuinely one of the dumbest things in electronics and I am convinced it exists purely to humble people who think they know what they're doing.
<img src="cables.jpg" alt="..." />
The board is called an ESP32-D0WD-V3. That's what they call it. I prefer to call it the thing that is eventually going to be inside my helmet, listening to everything I say, and routing it to an AI that knows my entire calendar, email, and active client engagements.
My fabricator is going to love this.
The sophistication is not in the hardware. The hardware is just ears and a mouth. The sophistication is in what happens between the two — a cloud pipeline that takes my voice, converts it to text, routes it through an AI brain connected to my Notion workspace, my Gmail, my calendar, my Airtable command center, and nine other tools, and speaks an answer back through the speaker.
All in under six seconds. On a $10 board. Inside a helmet.
<img src="esp32-board.jpg" alt="ESP32 board" />
<video src="detection.mp4" controls />
No great build comes together on the first try. I know this. I tell clients this. I still find it personally offensive every time it happens to me.
<img src="terminal-chaos.png" alt="Terminal chaos" />
HTTP status code 200 means "OK." It means the server received your request, understood it, processed it, and sent something back. It is the most ordinary number in web development.
Most people do not have feelings about it.
I sat back in my chair and did not say anything for a moment.
Because a board sitting on my desk — a board the size of a credit card, running code I had written, connected to a WiFi network named SpectrumSetup-9B — had found my server on the internet and knocked on the door.
<video src="first-200.mp4" controls />
or <img src="first-200.png" alt="First 200 OK" />
The board works. The firmware works. The API works. What doesn't exist yet is the actual helmet — the microphone inside it, the speaker behind it, and approximately twelve wires connecting everything together.
That part is next. I have a wiring guide. I have a fabricator. I have a build plan in progress that involves significantly more engineering than is strictly necessary, and that is exactly how I want it.
Three parts are on order. Combined, they turn a $10 microcontroller and a cloud API into something I can put on my head and have a conversation with.
The push-to-talk button already exists. It is the BOOT button on the dev board. I will hold it, speak, release, and wait. Then GRIOT will answer through the speaker.
We are one wiring session away from the first conversation.
<img src="wiring-guide.jpg" alt="Wiring guide" />
<img src="helmet.jpg" alt="The helmet" />
<video src="first-voice-test.mp4" controls poster="thumbnail.jpg" />
People ask me what I do. I tell them I help operators build AI infrastructure. That I design systems, not just prompts. That the difference between playing with AI and deploying AI is the difference between having a conversation and building a nervous system.
This is the nervous system.
It is a $10 board. It is a microphone the size of a fingernail. It is firmware I wrote at a desk — compiled, flashed, and confirmed working by a tool I control, running on a machine I own, connecting to an API I built. It routes my voice through a cloud brain that knows my clients, my calendar, my email, my entire operational context.
It is not a product someone sold me. It is not a subscription. It is not dependent on any single vendor staying in business, keeping the API free, or deciding my use case is still supported.
The helmet doesn't have a face yet. GRIOT does.
← Back to Wazirigaruba.com