Orbit — Proof of Personality Matching

Every social app matches you on what you claim to be — bios, hobby checklists, photos. But people describe their ideal self, not their actual self, so matches are made between two fictional people. That's why they die after "hey." Orbit takes a different path: Proof of Personality. Seven conversational questions about what you DID, not what you're like. Gemma 3 4B reads the behaviour in your stories and builds a profile where every trait carries a confidence level and is backed by your own words as evidence. Deterministic code guards enforce honesty: a trait only earns high confidence if its evidence quote actually appears in your answer. When Orbit gets you wrong, one click corrects it — you are the final authority on yourself. Matching runs as a three-stage funnel: hard filters (intent, dealbreakers — never overridden by score), embedding-based scoring, and an LLM rerank that names the connection type (mirror, complement, growth, opposite-world, shared journey) and explains the match in plain words — never a percentage. On mutual acceptance, Orbit writes the introduction itself and hands both people a concrete first activity: no more empty "hey." Everything — interview, extraction, matching, introductions — runs on a single AMD Instinct MI300X via vLLM/ROCm. The GPU interviewed and analysed 148 synthetic users in 2.1 minutes (0.8s per profile, 30 concurrent) — the same workload projects to ~40 hours on a consumer laptop. And because it's our hardware, the most personal data anyone can share never leaves the box. Privacy by hardware, not by policy.