30天內完成30個AI專案——數據揭秘

30 Days, 30 AI Projects: The Complete Retrospective

I shipped 30 production-ready AI projects in 30 days. Here's the honest breakdown: what worked, what didn't, and the real data behind the challenge.

Why I Did This

I left Meta after 8 years. 2B+ users. 25 engineers managed. Principal-level impact.

And suddenly, I was job hunting in the worst market in a decade.

The AI wave had changed everything. Every job posting wanted "LLM experience." Every interview asked about RAG systems and prompt engineering. My 18 years of backend engineering suddenly felt... incomplete.

So I made a decision: I would prove—publicly, undeniably—that I could build AI systems. Not tutorials. Not toy apps. Production-ready systems with tests, documentation, and real API integrations.

One project per day. For 30 days. No excuses.

The Raw Numbers

Technologies Used

Here's what I used across all 30 projects, ranked by frequency:

Project Categories

What Worked

What Didn't Work

The Daily Rhythm

Here's what a typical day looked like:

Key Insights

"Ship today" isn't a limitation—it's liberation. When you can't defer decisions, you make them. When you can't add features, you focus. The constraint created the speed.

Every day, the blank editor was intimidating. But once I had ANY working code—even just a function that printed "Hello"—momentum took over. Starting is 90% of the battle.

The core functionality? 2 hours. The error handling, retries, edge cases, rate limiting, graceful degradation? 6 hours. "Production-ready" isn't a feature—it's a multiplier on every feature.

18 years of backend patterns made this possible. I wasn't learning distributed systems AND AI—I was applying distributed systems TO AI. That's the difference between weeks and hours.

The Meta-Lesson

This challenge taught me something I'd forgotten after 8 years of management:

I can still build.

Not "I can still code"—I knew that. But building is different. Building is scoping, prioritizing, cutting, shipping, documenting, and then doing it again tomorrow. Building is the full stack of creation, not just the typing.

Management made me better at this. Watching hundreds of projects succeed and fail taught me patterns. Reviewing thousands of PRs showed me what separates good from great. Leading teams through ambiguity showed me how to make decisions with incomplete information.

The 30-day challenge wasn't a return to coding. It was a synthesis: Meta-scale production intuition + hands-on AI implementation.

What's Next: The Journey Continues

30 projects proved I can build. But building isn't the destination—it's the foundation. Here's where I'm heading next:

Going Deep, Not Just Wide

The challenge showed breadth. Now I'm developing depth in three areas:

Building for Real Users

30 projects with zero users is still zero users. My next goal: ship something people actually use. Whether that's open-sourcing a polished security library or building a small SaaS—real users force real quality.

Contributing to the Ecosystem

The AI ecosystem is moving fast, and I want to be part of building it—not just using it. That means contributing to open source projects, writing about what I learn, and sharing patterns that work. Knowledge compounds when you share it.

Staying on the Frontier

New models every month. Multi-modal becoming standard. Agents evolving rapidly. The only way to stay current is to keep building, keep reading papers, and keep implementing cutting-edge techniques. The learning never stops—and that's the part I love.

I'm not trying to become an "AI person." I'm trying to be a great engineer who happens to work with AI. The fundamentals—systems thinking, production quality, user focus—those don't change. The tools evolve, but the craft remains.

Looking for My Next Role

I'm actively searching for positions where this combination of experience matters:

If your team needs someone who can architect AND implement, lead AND ship—let's talk.

18 years building backend systems. Meta, Microsoft, Azure. Now building AI infrastructure.

Every project includes source code, documentation, and a detailed write-up.

Found this useful? Share it:

© 2025 Francisco Pérez Romero. Building AI tools that solve real problems.