When numerous companies began forcing large language models (LLMs) into every aspect of digital experiences, I began exploring the tools out of a longstanding curiosity about tech. Well, I also possess a pesky need to acquire employment in a society that has deemed it wonderful to threaten people’s daily survival if they stop producing Thingamabobs for a hypercapitalistic adult playground (“economy”) that has destroyed the planet by making fantastical assumptions about the relationship between said economy and the planet. In case it’s not apparent, I feel using LLMs is a very mixed bag. On the one hand, it accelerates all the dynamics hastening the demise of the latest iteration of civilization. On the other hand, LLMs can do some remarkable things, such as potentially unlocking discoveries that might slow down the destruction underway. What I don’t believe is this is just another tech like the printing press, the phone, or the internet. Those commodified distribution, and enhanced rather than replaced human outputs. LLMs are commodifying human creativity, analysis, and language, thus replacing (while also enhancing, sometimes) human outputs.
I mostly used Google Antigravity and Google Gemini, because they offered affordable deals and I presumed this massive company was well equipped to emerge successfully from the current competition among LLM (“AI”) services. I’m also invested in Claude, which impresses me as the most sophisticated tool out there for a curiosity-driven generalist like myself. Here’s what I built:
- Beshi**er: A game for people who like to create nonsensical words: https://bsgold.vercel.app/.
- CropZozz: A Mac app for rapidly cropping and exporting photos into any number of dimensions you specify at quality levels you desire. It has the extra benefit of allowing you to adjust the file names. I built this because I often need to prepare numerous photos for publication, and doing it manually is incredibly tedious. This app works well enough that I want to try selling it via the App Store. I am hesitant due to setting up an LLC for that purpose, which triggers tax complexity and cost.
- EventSnap: A Mac app that scans images or URLs to extract the dates, times, locations, and titles of events. I built it because I got tired of seeing event info on Instagram and laboriously updating my calendar. It has saved me an immense amount of time and gotten me to participate in more activities. I have not shared or sold the app. Not sure if I will.
- Know Your Friends: A game for groups of friends who want to see how well they know each other: https://know-your-friends.vercel.app/.
- Machina Tabellaris: An election results tracking app I built as part of a job application with Reuters: https://machina-tabellaris-git-main-hellotumos-projects.vercel.app/. I did not get an interview.
- Op-hed: People only read headlines now, so I figured I needed a way to address the many repeated failures of many mainstream media outlets to focus on the correct aspect of the disasters being inflicted upon the United States and world by a regime of seemingly suicidal-homicidal maniacs led by a convicted felon: https://ophed.news/. If you don’t speak journalism, op-hed (opinion-headline) is a play on op-ed (opinion-editorial). I built this via an elaborate back and forth with Gemini that allowed me to rapidly set up a custom theme and UI for a free CMS called Publii.
- TextsBuddy: A Mac app that remarkably enhances my ability to explore, organize, archive (if I want), and make sense of thousands of text messages written over years. I truly cannot believe I was able to build this. This is an app Apple should have built 20 years ago. I have not fully tested the archive functionality; it feels risky to rely on LLM code to protect data I value immensely. I have not sold it. I have not shared it. But I have used it quite a bit. Some of the most fun features: finding out how who texts you the most, viewing texts in a format like screenplays, tagging texts, organizing texts into folders, hearting specific texts, and otherwise fixing Apple’s ridiculous failure when it comes to exploring and managing old texts. I used Claude to clean up Gemini’s first iteration, and the performance got even better. I’d love to delete and archive old text messages, but it’s pretty risky given the software will eventually break as operating systems change, and I’d need to go through the slot-machine coding process again (vibe coding is just hoping and praying the machine produces the desired result after pressing the button with your questions).
- Truffle Pig: If you’ve purchased truffles from the Whole Foods bins and forgotten what flavor you chose because the wrappers are mysterious, suffer no more: https://hellotumo.com/trufflepig/.
- WordZozz: Another Mac app. When you conduct digital production for editorial organizations, you run into this horrific program called Microsoft Word. You need to clean up the text to remove weird line breaks, unusual styles, and insane URLs. I got very, very tired of doing this tedious work manually. WordZozz hooks into a remote LLM provider via an API or it uses your local LLM (Llama or LM Studio). The app also auto-generates tags, writes summaries, and produces SEO/GEO headlines. I generally am happy with the tags but unimpressed by the summaries and headlines. This project also taught me a lot (by using Gemini) about using Google’s AI Studio and setting it to avoid surprise massive charges.