Inspiration
We came up with the idea for Caught4K based on experiences from friends and families. There is a close friend that was interested in a person they met on an online dating app but something felt off for them. Before they went on a date they sent over an image that I immediately reverse image searched with Google but ran into an issue. Google has issues identifying people directly. This kicked off the journey to find a solution to the issue. We found an amazing service to identify people based on their selfies. That crude image upload resulted in some websites where the person posted images. It turns out that selfie was from an Instagram page froma completely different person. We caught the scammer in 4K.
These types of Romance scams using stolen selfies and and AI generated catfish profiles are exploding. The people most targeted young and older women dating online don't have a fast, private, judgment free way to check a match before things go further. Every existing "catfish detector" we looked at does one narrow thing (usually just AI photo detection) and none of them take the actual date into account. We wanted "run him before you meet him" to be a single verdict, not five different apps.
What it does
Drop in a photo, a profile link, or a screenshot of the DMs, and Caught4K hands back a verdict — CAUGHT, KINDA SUS, or LOOKS REAL built from three real signals instead of one: reverse image search , AI-generated/deepfake face detection, and message analysis (is this the "let's move off-platform," "my phone's broken" script every scammer reuses?). No account, no sign up, nothing kept the whole product is built around a "nothing stored, deleted the second we're done" promise that shows up on literally every screen, not buried in a privacy policy nobody reads.
How we built it
Caught4K is an app we built with Rork using Expo (React Native) app targeting iOS, Android, and web from one codebase, with a distinctive poster/zine visual identity (ink + paper duotone, Anton/Archivo/SpaceMono type system, hard-shadow "sticker" cards) that's meant to feel more like a printed zine than a generic pastel dating safety app.
For the detection engine, we stood up a small Hono backend deployed to Vercel that proxies requests to reverse image search, AI generated image detection, and text moderation — so nothing sensitive ever gets processed or stored client-side, and nothing is stored server side either; each request is scored and discarded.
For the Shipaton, we integrated RevenueCat to run the whole freemium layer. The user gets one free check, then a Caught4K+ subscription (weekly/monthly/yearly) that unlocks unlimited checks and full verdict detail handling entitlements and paywall state across StoreKit and Play Billing instead of us hand-rolling receipt validation.
Challenges we ran into
The biggest one was making sure "AI powered" wasn't just marketing copy. Early on, the detection engine was literally a deterministic hash of the input string. We it produced a confident looking verdict with zero relationship to reality. Replacing that with real AI calls, and discovering mid-build that the capability we needed actually lived on a different API generation (V3) than the public docs implied, took real debugging against live inputs until we could verify: a known AI-generated photo correctly scored CAUGHT, a real photo correctly scored LOOKS REAL, and a scam DM with off-platform-redirect language got flagged correctly.
Accomplishments that we're proud of
We were happy that the app was finally approved by Apple. Being in the Apple App store helps support our mission to help people not get scammed in the dating world.
What we learned
We learned how much of "trust" in a safety product is architectural, not just text copy. The privacy promise only means something because the backend truly discards every request instead of just saying it does. We got hands on with RevenueCat's entitlement model and how much complexity it removes from cross platform subscription state. We learned the hard way that AI vendor documentation can lag what their API actually does, so verifying real behavior against real inputs beats trusting the docs.
What's next for Caught4K AI Catfish Detector
Wiring the Safety Toolkit for real live Contacts integration, real location sharing during a Safe Date, and true SOS dispatch with legal review given the safety claims involved, plus resolving our one remaining paywall inconsistency so free vs. paid is enforced in exactly one place.
Built With
- expo.io
- noise
- revenuecat
Log in or sign up for Devpost to join the conversation.