Onboarding & intent signals
During registration we capture relationship intent (casual vs. serious spectrum), scheduling realities, and lifestyle anchors—pets, travel cadence, desire for children. These become structured signals instead of guesswork from a three-line bio.
Compatibility model (not just photos)
Our models blend explicit preferences with observed behavior: who you message, how long conversations last, and whether meetups get confirmed. Distance still matters—but secondary to value alignment for members who opt into “serious” tracks.
- Account created → device reputation + rate limits
- Profile enriched → quality score unlocks wider discovery
- Candidate retrieval → diversity constraints reduce filter bubbles
- Conversation → classifiers watch for harassment patterns
- Post-date feedback → tunes future suggestions (opt-in)
Human + AI moderation
Automated systems triage reported content and proactive signals (spam bursts, scam phrases). Human reviewers handle context: sarcasm, cultural nuance, contested boundaries. We keep audit logs for compliance and to reinvestigate appeals.
Safety releases before growth hacks
We delay launches that could spike risky contacts — for example, sudden global passport features — until reporting dashboards prove stability. That is slower growth, but healthier retention.