My 10-Month "Vibe Coding" Reality Check
I spent 10 months building and launching lupath.ai—a precision RAG travel discovery app. Here is the raw summary of what it actually takes to vibe-code a product from scratch.
The Spark: I picked up VS Code and GitHub Copilot. Seeing my first HTML/CSS frontend tweaks execute in the Terminal made me feel in total control.
The Reality Check: The honeymoon ended fast. Editing "borrowed" AI code is incredibly uncomfortable without syntax knowledge. Blind iteration is a trap; structure matters.
The Engine: I built a guided prompt funnel. To ground the LLM, I skipped messy web scraping and manually built custom JSON databases of 6,000+ ski resorts and global beaches to act as a RAG system.
The Identity: I fixed data-poor edge cases for tiny countries, bought the domain, and fed a notebook sketch to 5 LLMs. Gemini crushed the logo design. By late October 2025, development was done.
The Nightmare: I expected deployment to be a simple "Publish" button. Instead, Heroku slapped me with a memory-limit
SIGKILLerror because of my heavy JSON databases. After weeks of brutal troubleshooting, it finally went live on December 22, 2025.
Where It Stands (Mid-2026)
The app pulls about 150 uniques a day, but over 50% is just bot traffic. Human traction is low, conversions to Booking.com are minimal, and my building enthusiasm has hit a wall of marketing exhaustion.
The Indie Takeaway
Is it a financial goldmine? No. Is it a success? Absolutely. I proved that today, anyone can build a stable, functional AI app from scratch.
But as technical barriers crumble, the distribution wall grows taller. The future of indie hacking isn't about who can code—it's about who can market. Tech giants will make a killing on ads as millions of indie devs scramble to be heard.
TL;DR: AI builds the engine, but you still have to find the passengers yourself.
