Most dating apps optimise for scroll depth — infinite profiles, filtered photos, frictionless swiping. Laveyoo's founder wanted the opposite. The thesis: if users can't upload curated photos, have to answer real questions about themselves, and are rate-limited on likes, every connection becomes more intentional — and more likely to last.
Turning that philosophy into a shippable product meant solving four hard problems in parallel. Prove a user is real, not a curated highlight reel — no uploads, every photo captured live with face-liveness verification. Make compatibility quantitative via a 24-question assessment across four personality dimensions feeding a transparent scoring engine. Enforce scarcity — 5 likes per 24 hours, 72-hour like expiry, one active chat at a time — without breaking retention. And do all of it on a timeline that didn't require committing to a full custom engineering team on day one.
We scoped the build into five weekly milestones, each shippable on its own — so the founder could test real flows early, catch edge cases in staging, and re-prioritise without ever sitting on a dark branch.
Phases one and two delivered phone-and-OTP onboarding through Twilio, camera-capture selfie flow with single-face detection, and the database schema. Phases three and four built the 24-question compatibility engine with deterministic scoring, the matching engine with mutual-like thresholds and compatibility-weighted ranking, one-at-a-time chat logic, and the admin console for moderation, sweep reports, and priority debugging.
Where Bubble couldn't reach — specifically, sub-second video streaming to AWS for liveness detection — we shipped a dedicated React module. Amplify-hosted, Cognito-authenticated with guest credentials, streaming straight to Amazon Rekognition with an 85% confidence threshold. The Bubble backend stayed the single source of truth; the React sidecar is stateless, never touches the database, and redeploys independently. The Bubble workflow API is the only place that holds AWS verification secrets — the frontend never sees privileged keys.
Phase five wrapped the app as an installable PWA with service-worker caching, wired up the daily NY-timezone sweep scheduler that refreshes like quotas and runs match selection, and put the whole thing through QA for launch.
Laveyoo shipped to production inside the agreed eight-week timeline, with every scoped feature live and a clean handover of admin tools to the founder. The React-plus-Rekognition handshake clears liveness in a few seconds on typical 4G. Admins can inspect the priority list for any user and trace exactly why a match did or didn't fire. Daily-like quotas, 72-hour like expiry, and one-active-chat rules all hold up under concurrent edge cases.
Post-launch, the engagement continues in small, scope-boxed iterations — UI polish, sweep tuning, admin UX upgrades — as real users stress-test the product. The split between Bubble and the React sidecar means either side can evolve without touching the other.