Social Enzyme

Social Enzyme App

Inspiration

Loneliness and social anxiety among young people, plus the effort of organizing hangouts and finding people you actually click with. We wanted an AI that could represent you, find compatible people, and plan the day so you just show up.

What it does

The digital twin will find best matches for users, pinpoint shared interests, and even handle itineraries and bookings.

How we built it

We use Next.js, Supabase, and Minimax for voice + LLM; agent-to-agent conversations between AI twins do the matching, and we score compatibility from those transcripts. Itineraries and venue research use Exa and a Telegram bot for bookings, with Mem0 for persistent user memory.

Challenges we ran into

Minimax calls are slow, so running N×(N−1)/2 pair conversations had to be parallelized and tuned so matching didn’t time out. Making the “AI twin” feel accurate from limited onboarding (voice, social, or survey) and tuning the judge prompt for consistent compatibility scores were also tough.

Accomplishments that we're proud of

The real simulated conversations between AI personas instead of simple interest overlap, and the process of building the full loop from onboarding to UI Design to model training.

What we learned

Simulating social fit with AI conversations works but needs careful prompt design and scoring so groups feel right. Combining voice, social, and survey onboarding gives flexibility. We also concern about API latency and cost when scaling pair-wise matching.

What's next for Social Enzyme

We’ll refine matching (better judge prompts, group-balance logic) and add real venue integrations and booking flows so itineraries are bookable end-to-end. And we’re putting privacy front and centre: clearer consent and controls (what data we use for your AI twin, who can see it, and the option to delete or export it), plus transparent handling of voice and social data so users know exactly how their “second self” is built and used.

Longer term: more cities, recurring groups, and using post-trip feedback to improve both matching and planning.