I’ve been documenting technical projects for nearly two decades and though I enjoy the process, the fact is that it is usually more work than the project itself. Until you’ve done it, you don’t realize how difficult it is to solder while trying to record a nice video of yourself soldering. Repeat that for every step, then compile it all into a coherent and technically sufficient narrative — it is enough to make you hate the hobby. Kevin McAleer was tired of that, so he built this system that automatically documents his projects.
This seems impossible at first, because there is so much variation between one project and the next. How do you automate that?
McAleer realized that most of his project logs and tutorials follow a similar format. Each is a blog post that goes through the project step-by-step. It starts with an overview, then a list of components, and then each step of the build with accompanying photos. He concluded that if he could feed an AI all of the relevant information and material, the AI could pop out a blog post in the same format.
The breakthrough was in developing a way to give the AI the what it needs to concoct the blog post. To do that, Kevin McAleer selected an Arduino UNO Q. Its STM32 detects either a button press or soldering iron current, signaling that something is happening on the bench. A webcam placed over the bench captures visuals and its microphone records whatever McAleer says while he works.
As McAleer works on the project, the UNO Q passes all of the audio and regular video frames to Claude’s Sonnet AI model. The video frames give it visuals to select from and analyze, while the audio gives it detailed information. If, for example, McAleer says “I’m connecting the SCL pin of the sensor module to pin 22 on the development board,” Sonnet can make a note of that for the documentation.
When the build is done, Claude’s more powerful Opus AI model takes over to synthesize the information and visuals into a blog post.
McAleer can refine that and correct any mistakes made by Claude, but the bulk of the documentation work is completely automatic. For even a simple project, that can save him several hours of monotonous work.
