I Shipped 22 Projects in 2 Years. Here's What I Learned.
In the last two years, I've shipped 22+ projects. Four are live products. The rest taught me something worth the failure.
Here's what I've learned from building across AI, SaaS, infrastructure, and creative tools - all at the same time.
The projects
They range from mermaid.help (an AI-powered Mermaid diagram fixer born from daily frustration) to Laserman V2 (a multimodal AI film-production OS) to OneBiz (a multi-module SaaS platform wrestling with Indian GST compliance).
Some are one-weekend builds. Some have been in active development for months. Some are paused, waiting for the right moment.
What I learned
1. Build from frustration, not hypotheticals
Every project that shipped successfully started with a problem I personally had. mermaid.help exists because AI models generate broken Mermaid code. My remote TDD pipeline exists because I needed to code during train commutes.
The projects I built because "someone might want this" are the ones that stalled.
2. Self-hosting is a philosophy, not a cost optimization
I self-host OpenClaw, Kong, Keycloak, Novu, Getlago, and MeiliSearch on AWS. People ask why. The answer isn't cost - it's control. When you own your stack, you understand your stack. When something breaks at 2am, you know where to look.
3. The TDD pipeline changed everything
Running a full test-driven development pipeline from my phone via Claude Code + Infisical during train commutes sounds absurd. But it works. I ship features during my 45-minute commute that would have otherwise been dead time.
4. Shipping > perfecting
OneBiz has GST calculation bugs we're still fixing. Laserman V2's Chrome extension is janky. mermaid.help has edge cases. They all shipped anyway. Users found the bugs, and now they're getting fixed with real feedback instead of hypothetical scenarios.
5. Too many projects is a feature, not a bug
People say "focus on one thing." I understand the logic. But my best insights come from cross-pollination. The data modeling patterns from Laserman V2 influenced OneBiz's architecture. The self-hosting experience from OpenClaw made the notification stack setup trivial.
What's next
I'm doubling down on Laserman V2 and OneBiz in Q2 2026. The rest stays in maintenance mode. But I won't stop building new things - I'll just be more intentional about which frustrations deserve a full product.
If you're building multiple things simultaneously, don't let anyone tell you it's wrong. Just make sure each one teaches you something the others can't.