IntroMesh

IntroMesh is an in-person networking layer for events. Loop: Discover → Understand → Connect → Learn. Problem: at events you don't know who to talk to or why. Attendee lists show names, not reasons. Good conversations happen by luck. Solution: join an event, choose visibility (Active/Seeking, 90s expiry), browse a 3D signal field of attendees, read a grounded "why you should meet" explanation, and request a mutual connection. Share via link, QR, or vCard. Organizers only see aggregates with k≥5 anonymity, never private profiles. What makes it different: ranking is deterministic (goals×3 + interests×2 + skills×1 + complements×2 over max 24 candidates, top-8 persisted). Meta never scans the room — it only explains recommendations already chosen, with Zod-validated structured output. No invented interests: if the model cites data not in the profile, it's rejected. Without a key, the UI says so (503 meta), no fake text.