Thoughts Into Things

— I. The Pattern —

When I was seven years old, I helped my father build our house. Not metaphorically — we did the electrics, the plumbing, the brickwork, the roofing. He was a merchant seaman, home between voyages, and when he was home we built. That was my normal: you have a thought, you use your hands and tools and materials, and you turn it into a thing.

By ten I was obsessed with electronics. I ran wires under my bedroom carpet, drilled holes in the walls, secretly cut floorboards with the saw blade on my Swiss army knife to route cables and hide things. I connected wires to switches, buttons, speakers, microphones — and used them to make my voice come from places no one expected. I hollowed out my father's copy of Captain Hornblower, hid a radio transmitter inside it, placed it near my parents, crept to my room, and tuned in to eavesdrop. I'm sorry, Dad.

I loved James Bond. Enid Blyton. The Famous Five, the Secret Seven. Hidden buttons, secret passageways, invisible features. Magic. Not the supernatural kind — the kind you build. The kind that makes someone gasp because they didn't know the thing they were touching could do that.

That pattern — dream it, build the tools, find the materials, invent the missing links, turn the thought into the thing — has been the throughline of my entire life.

— II. The Education of Hands —

My family moved with my father's work. Oman, where at fourteen I stood on oil drilling platforms deep in the desert and drank goat's milk coffee with Bedouin families. Brunei, where at sixteen I worked alongside mechanics who had grown up in longhouses in the jungle. Australia, where at twenty I arrived alone with no money, no contacts, no plan, and found myself on a sheep farm in far western New South Wales — the landscape where the first Mad Max was filmed. I worked there for years. Twelve hours a day on a motorbike, chasing sheep, fixing fences and wells, learning under the constant ambient lesson that if you fuck up, you die.

I failed high school. I was confused, mostly misunderstood. But I could build things.

At twenty-four I went back to the UK and earned a degree in electronics. That led to a physics PhD at Cambridge, where I built tools — elaborate, precise tools — to pattern wires ten nanometres wide into circuits that, when cooled to liquid helium temperatures, could manipulate the flow of electrical current one electron at a time.

After Cambridge I joined Plastic Logic, a startup founded by Professor Richard Friend, and built inkjet printers that could print polymer transistors layer by layer. This was twenty-five years ago. After four years, I founded Novalia.

— III. Novalia: Twenty Years of Almost —

The goal was simple and absurd: use printing to create electronic circuits that could make magical experiences. Invisible technology embedded in everyday objects that would surprise and delight the humans who touched them.

Over twenty years, Novalia became the world leader in printed capacitive touch interfaces. Conductive ink on paper and packaging that turned physical objects into interactive surfaces. We worked with Ikea, Google, Bud Light, Amazon, Hershey's, Coca-Cola, Pepsi. We filed over forty patents. We invented missing links that didn't exist. We turned thoughts into things, at scale, for some of the biggest brands in the world.

But at the very start of Novalia, I transitioned. I'm a trans woman. And I found that no matter how much people loved what I built, no matter how many patents I filed, no matter how many Fortune 500 companies used my technology — there was always a safer pair of hands for the money to go to. I was an "almost happened." Twenty years of funding rounds that were always low six figures when they should have been millions. Always enough to survive, never enough to win.

We were pummelled by the pandemic. We came close, so many times. Just over a year ago, we ran out. My team of eight lost their jobs. Our investors lost interest. And I found myself in a studio just off Skid Row in downtown Los Angeles, with almost no money, often not enough to buy food.

— IV. Skid Row —

Every day I walk to my studio past people living in extreme suffering. Tents on sidewalks, people in crisis, the visible evidence of what happens when systems fail humans. It recalibrates you quickly. I know how lucky I am.

My girlfriend and I screen-print t-shirts using my printing skills. We sell them to organisations that support former drug addicts, helping to inspire them with the messages we design. That income funds the studio. It's not much.

I am extremely driven. All I know how to do is build. And I am guided by the signal of joy — the feeling that tells me I'm building the right thing, even when I don't yet know what it is.

— V. The Desert and the Cloud —

What pushed me over the edge was Govee lights.

I was in the Mojave Desert with my Govee LED lights, connected to my phone via Bluetooth. I wanted to change the colour. But the app required me to log into a server in China. I'm standing in the desert. The lights are right here. The phone is right here. The Bluetooth signal is travelling thirty centimetres. But somewhere between my finger and my lights, a server in Shenzhen has inserted itself to monitor what colour I choose and where I am. That server has no business being in this transaction.

My Renogy solar panels on my car — I can't check my power because the app wants me to log in. My Spotify music — I can't play songs I downloaded because they've been offloaded, and there's no signal. I'm in the Mojave. I should be able to do these things.

This is the reality of modern consumer technology. Every device, every service, every experience is routed through a cloud server controlled by a company funded by venture capital. The VC decides which products get built. The cloud decides when you can use them. The subscription decides how long you can keep them. This is not technology serving humans. This is technology extracting rent from humans.

So I built my own.

— VI. PiOps: A Sovereign Cloud in a Pelican Case —

It started as eight Raspberry Pi computers in a portable case, powered by a solar panel. Total cost: around seven hundred dollars. I didn't know where it was going. I just pursued my joy.

Each Pi costs between thirty-five and seventy-five dollars. They're single-board computers the size of a credit card, designed for education. My cluster runs its own DNS, its own file storage, its own wiki, its own AI inference, its own media server, its own home automation. No cloud. No subscriptions. No corporate dependencies.

The AI models run locally — fifteen open-weight language models on the cluster, plus a twenty-five-dollar Hailo neural accelerator that runs object detection at twenty-nine milliseconds per frame. Speech recognition and text-to-speech run on a Pi. The voice pipeline spans three Pis and six protocol conversions, entirely within my studio.

The internet connection is a Verizon hotspot. It's optional. The system works without it.

Over the last few weeks, working with Claude — an AI coding partner running in a terminal session on one of the Pis — I built what I can only describe as a sovereign creative technology platform. I didn't plan it. I built it one day at a time, making sure each day's work made tomorrow more productive.

— VII. What We Built —

The system is called Eidolon. It's a voice-first AI operating system that unifies every tool I need into a single conversational interface.

I can talk to it. I can type to it. I can use it from my phone, my laptop, or by speaking aloud in my studio. It manages a fleet of ESP32 microcontrollers — five-dollar chips that I solder to sensors, LED matrices, cameras, touch surfaces, speakers, and audio equipment. Each device has custom firmware that I develop conversationally with Claude: I describe what I want, Claude writes the code, builds it, and deploys it over the air to the device. A script verifies that the new firmware is running. I commit the change to version control.

The numbers, in three weeks of building: 24,000 lines of AI engine code. 121 commits to the Eidolon repository. 57 commits to the firmware repository. 12 device roles (LED art, touch MIDI, cameras, speakers, power monitors, audio players). 6 live devices on the network at any time. 113 session logs in the wiki documenting every step. 7 tool modules (fleet management, device studio, music, wiki, CAD, PCB design, LoRa mesh). Zero cloud dependencies for core functionality. Zero subscriptions.

The wiki automatically syncs to my reMarkable tablet. When I need to build hardware, I read the build guide on the tablet at my soldering bench while the cameras in the studio watch my work through the AI vision system. When the hardware is assembled, I plug it into the Pi and say "flash it" — the voice interface detects the new device, asks what role it should have, flashes the firmware, and registers it in the fleet. I open Device Studio, configure its MIDI channels and touch sensitivity through a panel that was built in conversation with Claude hours earlier, and test it in Ableton Live.

The distance between having an idea and holding a working interactive device is now measured in hours, not months.

— VIII. The Rania Moment —

On the evening of March 11th, my friend Rania — the artist Portrait XO — came to the studio. We soldered two LED displays from scratch: an 8x32 matrix and an RGBW strip. Components to working hardware in under an hour.

Then Rania, who had never flashed a microcontroller in her life, started talking to the system.

She asked for a pattern representing how Claude "feels." The system generated two sine waves drifting through deep blue-indigo and soft violet, with warm golden-white sparks blooming where they cross. A slow seven-second breath underneath. It deployed over the air to the hardware. The LEDs changed.

She asked for a climate data visualisation. NASA temperature anomalies rendered as brightness columns. Deployed. Running.

She asked for a tornado. A funnel sweeping across the display with debris particles orbiting with angular momentum, ground dust, lightning flashes. Deployed. Running.

Three animations, conversationally designed, deployed to physical hardware, running within minutes. By someone who cannot write code and had never programmed a microcontroller.

The system didn't write code for her. It wrote code with her. She described what she wanted. It showed her what it made. She guided it. The hardware changed. That's not automation. That's collaboration.

— IX. RYAT and the Wearables —

Today, March 17th 2026, the visiting artist RYAT flew in from New York. We built four wearable touch-MIDI instruments — capacitive touch surfaces that connect via Bluetooth to Ableton Live and turn the human body into a musical interface.

The build instructions went from the wiki to my tablet. I assembled the hardware at my bench, watched by the studio cameras. The firmware was flashed using the voice interface. Each device was configured in Device Studio — MIDI channels, touch sensitivity, slider assignments — through panels that Claude and I built this morning. We tested them in the MIDI visualiser, then in Ableton.

Four bespoke musical instruments. Designed, assembled, programmed, configured, and tested. In one day. By an artist with no engineering team, no funding, no cloud services, on hardware that costs less than dinner for two.

RYAT has now left to fly back to New York. The instruments work. The music exists.

— X. The Teeth —

In twenty years of running Novalia, my typical experience was this: I would have an idea — a clear vision of an experience I wanted to create. I would know about ninety percent of how to implement it. But that last ten percent required deep domain expertise I didn't have. So I would go to an engineer.

The engineer would do this thing with their teeth. They would suck in air and say: "Hmmm. I don't think you understand. That's not possible."

Well, it was possible. I would push, and eventually they would do it, reluctantly, and then tell me I got lucky this one time.

I spent twenty years fighting that dynamic. Expensive, often argumentative engineers — typically white, typically male — who treated my vision as naive because I couldn't write the code myself. Who confused their domain expertise with authority over what should be built.

Claude does not have teeth. We get along fine.

This is not about replacing engineers. It's about removing the gatekeeping that sits between creative vision and technical implementation. The bottleneck in hardware creativity has never been the hardware — ESP32 boards cost five dollars, LED strips cost ten. The bottleneck is the behaviour layer: what the hardware does. That has always required someone who can write firmware. Conversational AI collapses that gap entirely.

— XI. What This Means —

I did not set out to build a platform. I set out to pursue my joy and make sure each day's work made tomorrow more productive. What I ended up building — without funding, without a team, without enough money to reliably buy food — is a sovereign creative technology system that lets an artist turn thoughts into things at the speed of conversation.

This matters beyond my studio for three reasons.

First, the cost barrier is artificial. My entire infrastructure — AI inference, object detection, voice control, media streaming, firmware development, fleet management, surveillance art — runs on about seven hundred dollars of commodity hardware. No GPU. No cloud. No subscription. The technology industry has constructed an elaborate tollbooth between creators and their work. The tollbooth is optional.

Second, the expertise barrier is falling. Rania made three LED animations in an evening without writing a line of code. I built four MIDI instruments in a day without an engineering team. The last ten percent — the part that used to require an expensive specialist who would suck their teeth — is now a conversation. This doesn't eliminate the need for deep expertise. It eliminates the need to hire deep expertise just to explore an idea.

Third, diversity of creation matters. When only a narrow set of people — people with access to funding, engineering teams, and cloud infrastructure — can build interactive technology, we get a narrow set of experiences. When anyone with a vision and five dollars' worth of hardware can build an interactive device by describing what they want, we get experiences as diverse as the humans who dream them. The richer those thoughts and things are, the richer everyone's lives will be.

I have spent a year in deep poverty, building passionately in a studio off Skid Row. I did not know I was rebuilding my company. I did not know I was building a system for rapid deployment of deeply connected interactive media devices. I was just following the same pattern I've followed since I was seven years old: have a thought, pick up your tools, and turn it into a thing.

The difference now is that one of those tools thinks.