Suggestions, not assignments
Discovery tells people about plans and open seats that fit them. The host approves who joins.
The signal that matters
Mutual meet-again after a meetup is the main signal. It is private and mutual-only.
What it talks to
- API on Cloudflare Workers: One API for the whole product. It checks who the member is and what they may see, then calls the right module.
- Postgres with vector search: Stores profiles, hubs, plans, pulses and moderation outcomes, plus the embeddings used to find people and plans that fit.
- AI planner: Turns "something low-key Thursday near Main Street" into a draft plan with a few public places and times. The member edits it and posts it as the host.
- Mirror and onboarding: Gets a member to a first suggestion in about three minutes, then asks one question a day and shows what the system believes about them for them to confirm, correct or delete.
- Moderation pipeline: Classifies messages, photos, voice and reports, applies written rules through an action ladder, sends reasons and routes appeals.
Designed for
- Reported members drop out of recommendations at once
- No attractiveness ranking and no public scores
- Learning from behavior is opt-in
Where it appears in a request
- A report against a member: The reported member drops out of recommendations at once, pending the outcome.