AI Lab
Challenge
Game-theory-optimal preflop play lives in hundreds of dense solver charts, locked behind subscription web tools built for grinders, not for learning. I wanted a native Mac trainer that turned those static ranges into a fast drill loop, and as a test for myself: can I ship production-quality desktop software solo, with AI in the loop?
Strategy
I built the whole thing end to end with Claude Code and Gemini: a Python pipeline that parses raw solver range charts from CSV into structured JSON and programmatically fills gaps in the data, then a native macOS app in Swift that serves those ranges as an interactive trainer.
Results
A working native macOS app, shipped in two days, covering 9 table positions and millions of scenarios. More useful than the app itself, was continued proof that the distance between "idea" and "shipped software" has collapsed.