Rebuilt functionality from a $7,500/year structural-biology suite with Claude Code, and made it free

By now everyone knows AI for protein structure prediction got good enough to win a Nobel Prize. Things that were impossible a few years ago, computers just do now. But there's a weird gap here. The AI world talking about "we solved structure prediction" and actual wet-lab researchers getting to use that progress are two completely different things. Most of the state-of-the-art methods aren't built to be usable by researchers. You memorize command-line flags, you lose a day to setup, and nothing ever shows u