6 months and $485: My journey into building with AI
When I started building PicPickr, I was riding high on all the AI builder hype floating around online. Everyone seemed to be shipping things overnight. Given how simple my idea was (at least I thought so), I assumed I’d be done in a day or two.
I was so wrong.
It’s been six months now. Data, backend, frontend, Electron, CI, testing, landing page, design, Supabase, Resend, Lovable… the list goes on. What I thought would be a quick sprint turned into a marathon, and the todo list still doesn’t seem to end.
My plan in the beginning was simple: build something quick and dirty, and share it with other people. Something about that felt very human, very exciting. So I whipped up a web app overnight using Cursor. Then as I thought more about the problem, details emerged. I learned the hard way that web apps don’t play nicely with local file systems, especially when they need to handle thousands of large photos.
I then thought maybe an electron app would help with my objectives. Again, I think I gave AI too much credit and dove right in. I realized the real work had just begun. Packaging, distribution, signing binaries — all those invisible details that make something feel like a product. The funny part? I discovered these only toward the end, totally a surprise (not a good one, still waiting for my apple developer account approval) after months of building.
Even the landing page became its own project. I used Lovable to generate one, and soon enough, I was knee-deep in building flows for email signups, testing Resend integrations, tweaking copy, fixing layouts. It became another app in itself. If I were to do it again, I’d probably try something like Framer. I can’t even count the number of prompts I spent just refining that one page.
Building a good-looking UI is an underrated struggle. I ended up generating designs in Lovable and then manually re-creating them in my Electron app. Lovable has somehow cracked AI-driven UI generation, but integrating that into your real codebase is still a grind.
Most of my nights went into debugging Electron and its tools like Forge. At one point, I had to switch to Claude Opus just to get some specific config working. I was in config hell, and since I hadn’t written it to begin with, I couldn’t understand much or where it was falling.
After roughly 500 prompts, countless retries, and a lot of caffeine, PicPickr is finally live. At least the beta version. What is picpickr you ask?
PicPickr helps newlyweds sort through their wedding photos and pick their best ones for the album. It’s built for simplicity, with a swipe-based interface, autosave, and one-click export.
You can try it here: picpickr.com
What I learned
Over the years, I’ve had too many half-baked and abandoned projects. Ideas that started with energy but never made it to the finish line. This time was different. Being a systems developer with little experience in frontend or web tech, AI became my bridge. It made shipping possible. It kept me in rhythm. It helped me push through complexity and finish something that once felt out of reach.
If there’s one thing I’ve learned from these six months and $485, it’s that momentum matters more than mastery. Money ofcourse went in AI credits, I will share more details about it later.
I finally shipped something that exists not just in my head, but on the web, live and breathing.
Hell yeah!
Next up: figuring out marketing - distribution is such a pain.