Building simple applications by negotiating with LLMs

I'm a solo builder who enjoys building simple applications using AI in my free time. My setup includes Claude Code and Cursor with Swift on Xcode for iOS applications, Electron for PC, Railway for backend, Clerk for Auth, and Neon/AWS for PostgreSQL/RAG. I use API keys from Gemini, OpenAI, and Claude, and extensively use Midjourney, NanoBanana, V0, Lovable, and Figma AI. To build agents I use LangGraph, FastAPI, and Crew AI. To fine tune small models I've either used my own GPUs (for models under 8Bn parametres) or Replicate. Currently fine tuning a large model (Gemini 2.5) on custom data for a gaming application using Vertex AI.

I'm largely self-taught and frankly feel like a kid with a new toy when building with the AI tools. More projects fail than actually succeed, and while I have a few production applications, I dont have product market fit on any of them yet. But, not all those who wander are lost - and every wrong turn is a lesson.