Replace online outrage with real conversations across borders. Other Side takes a view you hold, finds three people around the world who see it differently, and hosts a calm, translated 10-minute conversation that ends with what you actually agree on.
2. Why people can't resist it
Being understood first. The AI steelmans your view back to you, better than you said it. People open up once they feel heard.
Curiosity about strangers. A nurse in Seoul and a cab driver in Lagos disagree with you. You want to know why.
The mind-meter reveal. Seeing everyone move a little, including yourself, is surprising and very shareable.
3. How it works for real
Voice in. Speech-to-text (Whisper large-v3 or Deepgram) on your 60-second take.
Steelman and position. A frontier LLM (Claude Sonnet or GPT) writes the strongest version of your view and places it on a stance scale with 3 reason tags.
Matching. A fast cheap model (Gemini Flash or DeepSeek) embeds reasons. We pick 3 people who disagree for different reasons, from different countries, with civility scores above a threshold.
The room. WebRTC audio and video (LiveKit), streaming transcription, translation by a fast LLM, captions in each person's language. The AI moderator (Claude) runs turns, asks one question at a time and pauses anyone who breaks the rules.
After. Everyone re-rates on the same scale. The LLM drafts the "what we agree on" card, and all four must approve it before it can be shared.
4. Cost per use
A 10-minute, 4-person room costs about $0.30 to $0.45 (4 audio streams of transcription about $0.16, translation $0.05, moderator LLM $0.06, relay $0.08). Free: 2 conversations a week. Plus ($5 a month): unlimited rooms, pick topics, keep transcripts. Schools, universities and newsrooms license moderated rooms for classes and readers.
5. Growth loop
The "what we agree on" card is the share unit: four flags, one topic, three shared points. It invites "I want to try that". You can also challenge a friend: "Record your take, I'll take the other side."
6. Risks and guardrails
Opt-in topics only. Launch list excludes elections, parties, candidates, religion-versus-religion and identity debates.
Moderator enforces rules live: no insults, no slurs, no personal attacks. Two warnings, then mute. One-tap report and leave.
First names and country only. No contact sharing. Recordings off by default and deleted after 24 hours.
No manipulation: the AI never argues for a side, never scores who won, and the mind-meter is self-reported.
18+ at launch, ID-light verification (phone plus selfie liveness).
7. MVP build order
Topic list, 60-second recording and the steelman screen.
Match on stance plus reasons from a seed group of 200 people across 5 countries.
Scheduled rooms with captions and a scripted AI moderator.