The problem I’m focused on is straightforward: understanding why a landing page doesn’t convert is more challenging than it should be. For engineers and founders, CRO analysis usually happens too late (after traffic, analytics, and experiments). For early-stage products, you’re often guessing what to fix or relying on generic advice that isn’t grounded in the actual page.

Crovise approaches this by analyzing a landing page directly. It ingests a URL, parses the DOM, extracts copy, hierarchy, and layout signals, and runs multiple analysis passes to surface concrete conversion hypotheses. The goal isn’t predictions or guarantees, but to generate testable ideas similar to what a CRO engineer might flag during a manual review.

Under the hood, it’s a combination of static analysis, heuristic scoring, and LLM-based reasoning. There’s no session data, no heatmaps, and no traffic signals yet — everything is derived from the page itself. This makes it fast and usable early, but also comes with obvious limitations.

I’ve been building this as a solo project and iterating based on real usage. I’d love feedback from engineers on the approach, edge cases, and whether this kind of analysis is useful without runtime data.

Thanks for taking a look.