Stay connected with KayaToday, follow us on Instagram and Facebook for the latest news and reviews delivered straight to you.
There is a particular kind of frustration that only arrives at 2am, when a power flicker forces you to stand in the dark jabbing buttons on a cheap clock-radio. For Lee Hutchinson, a writer at Ars Technica, that moment of irritation became the starting gun for a months-long project that ended up touching soldering, 3D printing, Python, Raspberry Pi networking, HomeKit integration, and AI-assisted CAD work. The result is a fully open-source, self-setting bedside clock with a red seven-segment display, and the journey to get there says something genuinely interesting about where hobbyist technology stands in 2025.
The finished project, including the bill of materials, software, and 3D printer files, is publicly available in a GitHub repository. But the more compelling story is not the clock itself. It is what Hutchinson had to assemble, intellectually and physically, to build something that the consumer market simply does not sell.
Why the Perfect Clock Does Not Exist on Any Shelf
Hutchinson’s requirements were specific but not unreasonable. He wanted a clock that sets itself automatically, handles Daylight Saving Time without manual input, stays accurate to sub-second precision, and shows the time on a red seven-segment LED display rather than the blue, green, or white variants that dominate the market. He also wanted no proprietary app required to make it work.
Battery-backed “atomic” clocks that receive radio time signals have existed for years and satisfy most of those criteria. The problem is that none of the ones Hutchinson could find combined radio synchronisation with a red seven-segment display in a form factor he found acceptable. The consumer electronics market optimises for the median buyer, and the median buyer apparently does not share his specific aesthetic preferences. That gap between “almost right” and “exactly right” is precisely where hobbyist projects are born.
His solution centred on a Raspberry Pi Zero W and later a Zero 2 W, chosen over an Arduino-style microcontroller because the Pi runs a full Debian-based Linux operating system. That choice brought Wi-Fi and NTP time synchronisation out of the box, along with the standard Linux remote management tools Hutchinson already knew. For the display, he chose an Adafruit 1.2-inch seven-segment LED module bundled with an HT16K33 controller backpack, ordering three units to account for the learning curve of soldering, which he had never done before.
Where AI Did the Heavy Lifting, and What That Actually Means
The software requirements Hutchinson set for himself were genuinely demanding. The clock needed to run as a systemd service under a non-privileged account, pull NTP updates from a local LAN server rather than the public internet, support scheduled display dimming, integrate with Apple HomeKit, and be deployable via a Gitea Actions pipeline so updates could be pushed without logging in manually. Most of the infrastructure work fell within his existing skills. The Python code required to communicate with the Adafruit display over I2C did not.
His solution was to hand the coding tasks to Claude, specifically Anthropic’s Claude Code tool running first on the Opus 4.8 model and then on the newer Fable model. He is candid about what this meant in practice. “Without the LLM, I wouldn’t have finished the project,” he writes. “I would have gotten annoyed, angry, or just tired of endlessly reading StackOverflow posts criticising what I’m trying to do for being dumb and wrong.” The LLM produced a tidy collection of Python files, a test suite, the HomeKit integration layer, the install routine, and most of the repository documentation.
This is worth sitting with for a moment, because it represents a genuine shift in what a technically literate but non-developer person can accomplish. Hutchinson describes himself as putting the “ops” in “devops,” meaning he understands systems and infrastructure but not software development. In previous years, the I2C communication layer alone would have been a hard blocker. In 2025, it was an afternoon of prompting. The LLM did not replace expertise so much as it bridged a specific gap between what Hutchinson knew and what the project required.
The same dynamic played out on the physical design side. Hutchinson found a Creative Commons-licensed 3D-printable enclosure built around the same Adafruit display, but it needed modification to accommodate smoked acrylic and neutral density filter material he added to tame the display’s brightness in a dark bedroom. Modifying the model meant working in Autodesk Fusion, which he describes with barely concealed hostility as an “absurdly user-hostile nightmare.” Fusion now ships with an MCP server, which allowed an LLM to remote-control the application directly. He first tried a locally hosted quantized version of Qwen 3.6-35B running on an Acer Veriton GN100 powered by Nvidia’s GB10 chip. The local model handled one of the two required changes correctly and failed on the other. Claude Code and Fable finished the job.
The Honest Accounting of a Hobby Project
The finished clock does exactly what Hutchinson wanted. It sets itself via NTP, handles DST through the host operating system, displays the time on a red seven-segment display at a brightness level comparable to a standard cheap clock-radio, and integrates with Apple HomeKit so the display can be controlled from an iOS device. The systemd service maps the Adafruit display’s 16 hardware brightness levels to a standard 0 to 100 percent slider in the Home app, with response described as essentially instantaneous.
The cost was higher than a shelf purchase would have been, inflated by false starts, a first-time soldering kit, redundant components, and multiple iterations of the 3D-printed case. Hutchinson does not provide a final figure but acknowledges it was “a lot more money than intended.” For anyone considering a similar project, he notes that Raspberry Pi Zero units are currently scarce, and suggests that an ESP32 microcontroller might actually be the smarter hardware choice since it includes Wi-Fi and I2C support natively without requiring a full Linux operating system.
What makes this project worth examining beyond its novelty is what it demonstrates about the current state of the tools available to technically curious people. The combination of affordable single-board computers, accessible fabrication hardware like 3D printers, component suppliers willing to sell in small quantities, and AI coding assistants that can bridge skill gaps has meaningfully lowered the barrier to building functional, well-engineered custom hardware. Hutchinson’s clock is not a rough prototype. It has a deployment pipeline, a test suite, HomeKit integration, and a documented installation script. A project of that completeness, built by someone who self-describes as not a developer, would have been substantially harder to finish even five years ago.
The deeper point is not that AI wrote the code, but that the combination of available tools now makes “build it yourself” a genuinely competitive answer to “I can’t find what I want on the market.” For makers, engineers, and technically minded hobbyists across the region, that threshold shift is worth paying attention to.
