Cross-platform discovery for iOS AI apps — semantic + vector search via MongoDB, AI-summarized descriptions, similar-app suggestions, bookmarks, trending keywords, and a quiz that recommends 5 apps from a few quick questions.
App Store search is bad for niche categories like AI tools — you either get the same five apps everyone already knows, or two hundred results with no signal. I wanted to find AI apps the way I search Google: by what I want to *do*, not by exact keyword.
Two builds in one — an iOS app and a marketing site — connected by a small backend. The interesting bit was semantic search: I stored each app's description as a vector embedding so a search like "summarize my PDFs" could match an app listed under "document analysis," even with zero shared keywords. For non-technical founders curious about adding AI to your product: the hard part isn't the AI itself, it's the boring data pipeline that feeds it. Most of my time went into keeping the catalog fresh, not into the smart search everyone wanted to talk about.
Two honest lessons. First: keeping the catalog fresh meant a backend job that broke constantly whenever the App Store quietly changed its data. Second: discovery products are brutal because nobody opens them daily — once a user finds the app they were looking for, they're gone. The retention curve told the story before any feature could fix it.
Shipped, learned, then sunsetted after a few months of honest numbers. If you're a founder considering a discovery or directory app, the build is the easy part — the hard part is convincing anyone to come back twice. I'd build it again with these lessons, but only with a clear hook for repeat use baked in from day one.