Skip to Main Content

MuseMeet

Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote

MuseMeet — Demo video

Millions of singles suffer from swiping fatigue, deceptive profiles, and superficial algorithms that treat human connection as an infinite slot machine. Empirical dating app archives reveal an alarming drop-off: out of 14,250 swipes, an average user sees only 118 matches, 46 chats, and 4 first dates, culminating in zero long-term alignment. Over 91% of conversations fizzle into ghosting, exhausting emotional bandwidth on repetitive interview-style small talk. MuseMeet introduces Sovereign Agentic Dating. Instead of humans wasting 10 hours a week swiping, personal Meta Muse companion agents negotiate romantic compatibility on their behalf inside a private, AirGap-protected local lounge, only introducing humans in the real world when deep, verified alignment is proven. Key Architecture & Innovations: 1. Sovereign Footprint Distillation (M1): With explicit user consent, personal Meta Muse companion agents distill authentic digital footprints (reading history, conversational tempo, chronotype, attachment values) into private 128-dimensional behavioral vectors with zero cloud egress. 2. 128-Dimensional Metropolitan Funnel: An on-device 5-stage retrieval pipeline evaluates candidates across 7 relational domains (Emotional Core, Lifestyle Rhythm, Attachment Security, Intellectual Drive, Communication Dialect, Spatial Proximity, Long-Term Values). Benchmarked at 8.77ms for 2,500 candidates, far exceeding the 35ms latency budget. 3. Autonomous Mascot Rendezvous (M2): In the AirGap Autonomous Lounge, two companion Muses meet across 4 empirical stages: Discovery, Banter, Friction, and Synthesis. As they converse, real-time Ireland-Pennebaker Linguistic Style Matching (LSM) tracks function word synchrony alongside Gottman bidirectional bid responsiveness. Hidden subtext monologues evaluate boundaries at machine speed. 4. Anti-Sycophancy Neutral Moderator (M2b): An independent arbiter continuously tests for false consensus, confirming whether agreement reflects authentic relational alignment or conversational appeasement. 5. Explainable Offline Dossier & Meta Moodboard (M3 & M4): When mutual alignment passes threshold, MuseMeet synthesizes an offline first-date itinerary grounded in local spatial venues, accompanied by a coordinated 3-panel visual moodboard (Atmosphere, Activity, Moment) generated natively by Meta Muse Image 1.0. 6. AirGap Contextual Integrity Defense: Kernel-level regex and semantic boundaries prevent prompt injection, intercepting raw PII, photos, or GPS data before it can leave the device. MuseMeet transforms dating from an exhausting algorithmic slot machine into intentional real-world human connection.

Team