AI-Assisted Projects

The seed was a corporate walk challenge at Prudential that stuck in my head. Years later I tried a 100-day challenge with family and a friend, tracked on a Google Form. It worked, barely. Two failed Lovable attempts later, I rebuilt it properly in Claude Code — and kept adding more than just walking. Yoga, meditation, Strava sync, swim, fasting timer. The certificate generated when you finish a challenge is the feature I'm most proud of. I use it every day.

Around the time I was getting really good at CSS, a product leader I worked with needed an interactive mobile demo — something polished, clickable, that felt like a real app. I built it as a side project, half genuine learning, half quiet daydream about selling it to him on my way out the door. One potential customer. Weeks of work. The AI tools of today would build the same thing in an afternoon. I got bored before I could pitch it. Recently brought it back to life.

I had Google Docs and Notes but never liked the friction of formatting before I could write. I wanted to open something, pick a category, and type. The goal: a private journal with categories, streak tracking, and local encryption. Named after the Kannada word ಲಿಖಿತ (Likita) — "written." Of everything I've built, this is the one I actually use most consistently. That's probably the best measure of whether something worked.

A member of my state's Kannada-language WhatsApp group writes a thoughtful piece on heart health every week — good content disappearing into the scroll. I wanted to give it a proper home: searchable, with a glossary and PDF. Built in a single session. What I learned: no one reads a bulk archive. The content was made for a newsletter, one piece at a time. The format you use to share something matters as much as the content itself.

Started as a simple addition app on Lovable, then built multiplication tables separately in Claude Code, then combined them. My daughter Arya uses it on and off — she's the only user. I want it to eventually save progress, adapt to her level, and let a teacher see how a class is doing. Added multiple choice, 3rd grade math topics, countdown timer, timed drill. Tried a Khan Amigo-style AI coach — the output was terrible and I disabled it.

I was using two separate apps — one for food logging, one for fasting. I wanted one that did both, plus things always locked behind paid plans. Started in early 2025, gave up in October when Lovable couldn't fix its own errors. Came back months later and Lovable had improved dramatically — the same bugs resolved in minutes. Still has one passionate user. Paused it because StreakMates kept pulling my attention.

Came across this ancient Indian board game about two years ago and wanted to build a digital version. The hardest part was the board — the geometry is specific and the game breaks if it's wrong. I went through several failed designs before getting it right, and kept the interesting failed variants in the app. Once the mechanics worked I lost interest in the polish. The UI could use work. But the game plays.

Two people in my life — my wife and a friend — independently asked for the same app: recipe recommendations from ingredients already in the fridge. I also found myself asking ChatGPT for cocktail variations constantly. Built one app for both needs. I knew from the start it was an AI wrapper — and sure enough, couldn't find a way to make it more than that. Close to a one-shot build. That's probably fine.

In my pickleball group I post a poll every week — takes 30 seconds, everyone can see who's in. Without me, the chat turns into 40 messages and nobody knows who's playing. I've been the only one doing this for 12 months. Built a Doodle-style grid: pick your availability, share a link, see where everyone overlaps. No signup. Rebuilt the UI seven times. The harder problem isn't the design — it's adoption. Anything that requires leaving the group chat is already losing.

I've had a blog for over 23 years. At some point the free host stopped working and the content became inaccessible — but I kept paying for the domain because I couldn't let it go. Rebuilt it with Claude: parsed a 70MB WordPress SQL dump, recovered 700+ posts across four different blogs, and built a static site served from GitHub Pages. The domain finally earns its keep.

My wife was finishing her data analytics program and needed a portfolio site. Built it on GitHub Pages — clean, fast, no backend. A small project but a complete one: it does exactly what it needs to do and nothing more.

At Prudential during COVID, I noticed something odd: the retirement tool asked users for their monthly expenses, but those expenses had no effect on the final recommendation. The field existed. The data went nowhere. I built a better version — three spending models, all reverse-engineered from scratch without touching the original codebase. One of the most fun moments was confirming that the expense field truly had no bearing on the output. It was also my introduction to React. Revived with Claude and published here.

[Your story here — what prompted this, what you learned, what surprised you]

[Your story here — what prompted this, what you learned, what surprised you]

[Your story here — what prompted this, what you learned, what surprised you]