In 2026 I started getting more serious about photography. In the past, I had posted my photos to Instagram, usually during some trip. The descriptions got longer as I found people liked reading about my travel. But I hated relying on Instagram, and I wanted more control over the quality and metadata of the photos. So I decided to build my own photo website, which I could link to from Instagram and Bluesky.

There’s no doubt AI has changed programming, and my photo site is a perfect example. I still love programming, and I have no reservations about getting into the weeds when I need to. But for a project like this, I was more interested in the design and outcome of it than actually implementing it myself. So for the most part, the code was written with Claude in a tight design feedback loop. This sometimes did require looking under the hood to make sure things were done correctly, but overall it was a pleasant cycle of design, test, loop. The speed of AI code allows you to experiment more ambitiously.

The look of the site was inspired by my film camera, the Pentax MX. I found fonts that best matched the Pentax logo (Archivo) and the copy in the manual (Univers). The “MX” logo on the camera is one of my favorites, so I spent a good amount of time reconstructing it in Affinity Designer and adapting it to my own “ZM” logo.

One of the first things I added was simple analytics. As much as I hate the invasive ways most websites track users, I recognize the need for some feedback for site owners. I was most interested in how many users saw each post, and how they got there. The simplest, least invasive method I could think of was leaving an anonymous visitor ID cookie on each device to avoid double-counting views. Links I post in different outlets carry a tag to let me know when visitors use those links. The result is a collection of graphs that help me see how users are using the site without tracking or identifying them.

An early benefit of those analytics was discovering that most users viewed the site from a mobile device (as expected), but that users didn’t realize the mobile view had a photo carousel they could swipe through. This led to some iteration on how to present the carousel in a more intuitive way.

This is a little experiment to simulate creating notecards with a typewriter. The card moves around as you type instead of the cursor, and you can’t undo mistakes—characters are typed over each other. I wanted to recreate the fun and frustration I had while making recipe cards on an old typewriter I got a while back. I spent a while trying to get the look of the typing indicator bar right, and I like how it turned out.

Over the course of 2023, I made a simple programming language from scratch in C, called Cassette. I wanted something simple and focused that I could use for small programming experiments and art projects. I designed the syntax and implemented a parser, compiler, and virtual machine for it. The syntax is inspired by Elixir, but evolved over the project. The most useful part of the project was the dynamic memory system, which implements the basic types and operations. I used this in the parser and compiler for intermediate data structures until late in the project, and the dynamic typing greatly sped up experimentation for those.

I wrote Cassette in C, mostly for performance and because I wanted the option to port it to an old Macintosh SE/30 I have. Writing in C as a solo dev on a side project has been a lot of fun, and very instructive, but I think I prefer higher level languages like Elixir for professional work.

There’s still plenty of future work and bugs to fix, but I’m pretty happy with the result, and I’m looking forward to using it and improving it in the future. For now, though, I want to turn back to Elixir and the Web.

At Zillow, employees can earn badges on their profile for various accomplishments. Many of the badges look really nice, because they were created by our excellent design team. However, badges are pretty low priority, so often they don’t have time to create new ones, and many badges are designed by others without quite the same polish.

The inconsistency started to annoy me, so I decided to make this tool that would help anyone create a decent-looking (or at least consistent-looking) badge. Most of the badges follow a pretty standard template, so I distilled the controls down to only what varied between them.

The color picker users Vibrant.js to select a palette based on the image you upload, to make it even easier to match colors. The settings control an SVG image, which is downloaded as a PNG when the user is finished. They can then submit it to an admin to add to the system.

Brutalist web design has always been a favorite of mine. I love the clean lines, the lo-fi vibe, and the dark alternative proto-Internet that we all like to imagine used to exist. In some sense, it’s liberating to demand that the user bend to the idiosyncrasies of the design rather than cater to what users want or expect.

This specific project was greatly inspired by the chat mini-game thing in Super Hot. Basically, I stole the whole premise and turned it into a website. I loved the idea of real-time text, where you could see people make typos, and wouldn’t have to wait staring at a “John is typing…” status. The automatic pseudonyms were also a fun trick, which I imagined would give some illusion of privacy, when in reality you can learn to distinguish people pretty quickly based on their typing speed and accuracy.

The server is written with Express, a popular Node.js framework. It’s pretty simple — basically just rendering a React app and setting up Socket.io to relay messages. The socket file supports a few commands, as well as basic messaging, and is also relatively simple; in fact, I think a chat application is one of Socket.io’s example apps in their documentation. I use Redis to keep track of the number of users online in each room, and to record conversations (when that’s enabled). The client side is a pretty simple React app, with Redux to store the messages. As with most side projects, the code could be factored out a little better, but given the size of the project the organization doesn’t seem too hard to navigate.

Showing the app to some friends, we all got a kick out of playing around with it for a while, but understandably it never replaced our normal messaging platforms. Occasionally a few people will pop in from time to time, but it’s not as interesting if you’re the only one online. It was also included on brutalistwebsites.com, which is about all I was looking for anyway.