Feelink

Affective neuroscience is clear about something most consumer apps ignore: emotional granularity — the ability to identify, label, and distinguish between emotions — is a stronger predictor of long-term wellbeing than emotional positivity itself. People who can tell the difference between 'frustrated' and 'overwhelmed' recover faster from stress than people who feel both as 'bad'. Yet most tracking apps reduce all of this to a single mood slider. Feelink was built to capture emotion the way people actually feel it: in the body, in context, with reasons.

The entire product was built in Lovable — Lovable's web editor handled the React + Vite + Tailwind + shadcn scaffold and every iteration after it. The git history (374 commits across 7 working sessions over two months) tells the iteration story: a body figure that took ten visual rounds to land, a click-to-place dot interaction that needed nine follow-up bug-fix prompts, an icon library swap from lucide to Phosphor that broke imports twice, and a five-revert-in-four-minutes slider styling sequence that became the test case for the design system. Every screen below is captured directly from the live Lovable app at feelink-app.lovable.app.

There are two AI layers here. First, in the build: Lovable was the entire dev environment. I spoke to it in natural language for every component, every refactor, every styling decision — the 374 commits in the git history are the receipt. Second, in the product: Claude Sonnet runs at two specific moments — suggesting body-sensation language when users get stuck describing where an emotion lives physically, and writing the weekly synthesis that connects emotional patterns to triggers and behaviors. Both of these earn their place because the alternative is staring at a blank field, and blank fields are where check-ins die.