What I've Been Working On Over the Past Month
Background
- I studied economics at university and later worked in consulting. I have no software development background.
- I can do basic things like buy and manage my own domain and build a simple company website with Google Sites.
- In the past, I tried Codecademy a few times, but I always gave up before I actually built anything.
- I have written small Python scripts before—less than 100 lines or so—to pull data through APIs and export CSVs for Power BI.
- But even then, I eventually hit a wall and quit.
- So to be clear, I’m really not a software engineer.
Things I Built This Past Month
Grouped by platform:
Vercel only
- An app that shows snapshot data from the Bitcoin options market
Vercel + Supabase
- A simple workout progress tracking app
- An app that recommends Japanese sake based on answers to four questions
Vercel + Railway
- An app that tracks Bitcoin momentum using different indicators
- An app that recommends wine based on three tasting tags
- An app that uses an LLM to process U.S. companies’ IR filings and generate summaries
- An app that runs Monte Carlo simulations to estimate portfolio survival probability based on safe withdrawal rate input
Cloudflare
- This blog — new!
How I Build
I use an M3 iMac, VS Code, and GitHub Copilot. It only has 16GB of memory. It’s a 2023 model, and back then 16GB was supposed to be enough—or at least that was the idea.
Usually, I start by talking with tools like Perplexity or Manus. Sometimes I begin with a vague question that I want to explore, and when I realize I can’t research or test it in a satisfying way, I end up building an app instead. The Monte Carlo simulation app came out of that process.
Other times, I start with a clearer goal: I want to build a specific thing. That was the case with the workout tracking app. I had originally been logging my workouts in the Notes app on my iPhone, but it felt clumsy. So I shared those notes with Manus, had it draft a product spec in markdown, and then used GitHub Copilot to build and implement the app from there. It was probably the first relatively complex app I built—at least by my standards—because it included not just a frontend, but also authentication and a database. In that sense, it was a great way to learn the overall flow of AI-assisted app development.
Reflections
Looking back, I’ve made quite a lot. What’s interesting is that because I have no engineering background at all, I have almost no instinct for how much work something is supposed to take. In a strange way, that helps: I don’t hesitate too much, and I just try building things. If I wanted to call it vibe coding, I may be a bit late to the trend.
The more things I build—or even just consider building—the more I feel that a good dataset is a real asset. That may sound obvious by now, but with apps like the wine and sake recommendation tools, I’m basically stuck on how to expand the database and improve its quality. On the other hand, one area where AI has made life much easier is content generation for structured lists. For example, if I ask it to produce a broad, reasonably comprehensive list of workout routines, it can give me something like 50 useful menu items right away.
The way I use my computer has changed completely. I never expected to spend this much time staring at VS Code. That said, the actual coding happens at high speed on the AI side—my latest partner has been Opus 4.7—so most of the time I’m really just watching it work. I’ve also gotten much more comfortable with the terminal. Still, I’m very much in a state where the AI is carrying me. So I want to keep building more things and deepen my understanding as I go.