AI Experiments

Things I've been building with AI lately.

Balancer

Finance tracker that helps you track what you earn, spend, and save so you can see whether you are on track toward your goals.

Cue

Beta

Open source desktop app that uses AI to create subtitles you can edit, style, and burn into your video.

My vibe coding process

I use Cursor a lot, but I do not hand it the wheel. It helps me move faster. I still make the calls on scope, product direction, quality, and what is actually ready to ship.

Every repo has its own roadmap, handoff notes, and product decisions. When a session starts getting messy, I stop and write down where things stand so I can pick it back up later without losing time or repeating the same mistakes.

How I work with AI

Workflow

1

Start with a real problem

If it is not affecting my own life, I usually do not build it.
2

See what already exists

I look at the current options and figure out where they fall short.
3

Decide the scope early

I write down the important calls before I start building.
4

Build with Cursor

I work in small passes so I can keep the product stable.
5

Leave clean notes

Roadmaps and handoff docs help the next session start fast.
6

Test before shipping

I use the product myself before I call anything done.

Most of these projects start because something in real life is annoying enough to keep bothering me. Balancer came out of money stress at home. Cue came from my wife needing good subtitles without paying for software or uploading private files to the cloud.

Once the problem is clear, I look at competing tools, decide what matters, and cut the rest. Then I build in milestones with Cursor and keep the notes current, so each new session starts with context instead of guesswork.

I do not ship something just because the code runs. I click through it, use it for real, and pay attention to where it feels awkward. If a feature works on paper but feels wrong in practice, I would rather pause it than pretend it is finished.