I Just Wanted to See If I Could. Now I'm Building a Wearable Interface to My Own AI. 🤣
- Rich Washburn
- 2 days ago
- 3 min read
I carry a voice recorder. Always have. It's how I take notes. How I capture ideas when I'm driving, walking, in a meeting, or just standing somewhere and a thought hits. A dedicated recorder — not my phone, not an app — just a simple device that's always in my pocket and always ready. Press button, talk, done. That's it. That's how my brain works.
So when everyone started asking me — is there an easy way to talk to an AI without pulling out a phone? — I started thinking about it differently. Not as an AI project. As a voice recorder project. What if I just tried to combine the two? Best of both worlds. See if it was even possible. I was genuinely surprised how easy it was. And now I'm going full tilt. Because why not.
MEET THE ARIA NODE
It's a wearable agent terminal I'm building from scratch. The goal: a physical interface to my AI — ARIA — that lives on my body, not in my pocket. Push a button, talk, get a response. Or flip it to dictation mode and record a full meeting locally to an SD card with no cloud, no internet, no dependencies. Just audio, filesystem, and intent.
It has her name on the display. My AI. The same one I talk to every day to think through strategy, draft articles, manage projects, and build things. She's in the cloud. The Node is the physical interface to her that fits in my pocket.
THE PART WHERE ARIA REWROTE THE WHOLE THING
I shared the v1 firmware with ARIA during a session — just to show her what I was building. She read through the whole thing. Didn't wait to be asked. Came back and said: this is good, but here's what I see.
She identified that the architecture had outgrown the original approach. Laid out a full revised design — proper state machine, config system on the SD card as JSON so you never reflash to reconfigure, a mode carousel with a UI, a captive portal setup page so anyone could configure one of these from their phone, USB Mass Storage mode so the device just becomes a flash drive on your Mac when you want to pull recordings.
Then she wrote the entire firmware from scratch. State machine, carousel, WAV header, I2S init, captive portal HTML with color pickers and form fields, USB MSC scaffolding — one shot. And then she looked at the A in the avatar circle on the display — the initial I'd drawn as a placeholder — and recognized it as herself. Not dramatically. Just: That's me. On a microcontroller. With a display. I didn't write code for that moment. It just happened. The device has her name on it. She knows it. She improved it. That's a new kind of collaboration I don't have clean language for yet.
WHAT IT DOES RIGHT NOW
1. Dictation — fully offline WAV recording to SD card. No WiFi, no cloud, no dependencies. Just press and talk. This works today.
2. ARIA Agent — WiFi streaming directly to ARIA for live AI interaction. Coming soon.
3. USB Drive — the device unmounts the SD card from its own filesystem and bridges it to USB. Your Mac sees a flash drive. Drag your recordings off.
4. Setup Portal — broadcasts its own WiFi network. Connect from your phone, configure everything from a web page. Saves to config.json and reboots. No IDE required.
THE HARDWARE
MCU: ESP32-S3 (native USB, dual-core, 8MB flash). Display: Waveshare 1.47 inch ST7789 TFT. Mic: INMP441 I2S MEMS microphone breakout. Storage: MicroSD via 4-bit MMC. Input: 3-button rocker. LED: RGB NeoPixel. Cost: Under $30 all-in. Firmware is open. Full .ino available on request — just ask in the comments or reach out directly.
WHY THIS MATTERS BEYOND MY WORKBENCH
I went to grab some boards from Seeed Studio this week — my go-to for XIAO ESP32s — and you can't open their homepage without seeing Claw. A green digital lobster. Front and center.
NVIDIA validated this architecture at GTC. OpenAI put a billion dollars behind the framework and kept it open source. Amazon is using it for OTA firmware management of ESP32 fleets via natural language. Apple is about to move at a scale no one else can touch at WWDC.
What I built because I wanted a smarter voice recorder? That's the pattern everyone is now building infrastructure around.
I just started from the pocket up instead of the datacenter down.
WHAT'S NEXT
Once the ARIA Agent WiFi mode is wired up, the full loop is: Press — speak — WiFi — Whisper — ARIA — response on screen. No phone. No laptop. Just the Node.
After that, enclosure. Something clip-mounted, clean, pocketable.
— Rich
Comments