I'm a dad of 2 who works night shifts I vibe coded this application with zero coding experience
Hey everyone,
I wanted to share the build journey of my first deployed app, because the pipeline I ended up using to get it across the finish line as a non-coder felt a bit absurd, but it worked.
The Project: I built Dermia (https://dermia.up.railway.app) — it’s a fast, free skincare routine builder that checks user inputs against 18 specific allergens and product-ingredient interactions to build safe AM/PM routines.
The Multi-Model Chaos / How I Built It:
- The Mockup: I started out "vibe coding" in Google AI Studio using Gemini. It was great for the initial frontend layout, but as the codebase grew and technical logic got deep, Gemini started suffering from massive context drift (forgetting components we established 10 prompts ago).
- Going Local (and hitting a wall): I switched to running Ollama locally to keep building, but running heavy local models for a complex web project wasn't sustainable on my hardware.
- The Breakthrough: I discovered Ollama has cloud models, which opened up free Claude models for a bit. But I hit massive friction trying to get Qwen to handle Claude's tool-calling logic correctly.
- The Final Engine: I bit the bullet and upgraded to Ollama Pro ($20 for 50x usage), which gave me access to GLM 5.1. I used GLM 5.1 as my primary "senior dev" to fix the tool-calling bugs, handle the database schemas, and fine-tune the allergen-checking logic.
Where it stands now: The app is live, fully functional, and deployed on Railway. It’s completely free, and you don’t even need an account to test the quiz logic.
I need brutal code/product feedback: Since I have exactly zero formal coding experience, I know there are things here that probably look insane to a real developer. What feels broken? What UI cues instantly scream "AI-generated" to you? If you test out a routine, does the ingredient filtering logic actually hold up?
Hit me with your worst, I want to make this genuinely good.