HiveMindGroup ideation for hackathons, fused by AI.
01
Where the idea came from
HiveMind came out of hackathons. I have sat through enough of them to know that the clock is the real opponent, and a surprising amount of that clock disappears before anyone writes a line of code, in the brainstorming. Everyone has an idea, everyone takes a turn explaining it, the team debates, and an hour is gone before you have converged on anything. With so little time in a hackathon, that phase is expensive. HiveMind compresses it. Everyone drops a short pitch, AI pulls the most distinctive element from each one and fuses them into brand-new concepts the whole team owns, and you get to building instead of debating.
02
How it works
The host creates a room and gets a shareable link, a QR code, and a six-letter code. Each teammate opens it and submits their own idea into a live pitch feed powered by Supabase Realtime. The moment the last person submits, the AI auto-triggers, or the host can force it early. It reads every pitch, extracts the single most distinctive element from each, and generates four fused project concepts, each credited according to the privacy choices the contributors made.
03
The privacy model
Every participant gets two independent toggles at submission time, hide name and hide idea, which combine into four levels from fully open to fully incognito. A fully incognito pitch still silently shapes every generated concept but earns no visible credit. The important part is that privacy is enforced server-side, not just hidden in the UI. The ideas table has no browser read access, every read flows through an API route that masks the per-row flags, the AI never receives a hidden participant's real name, and results are masked before they are ever stored anywhere the browser can read.
04
The stack
Next.js on Vercel for the app and API, Supabase for Postgres plus the realtime pitch feed, and a two-model AI setup with Google Gemini as the primary and Groq running Llama 3.3 70B as the fallback. Every layer sits on a free tier, and the automatic failover means that if Gemini rate-limits during a busy event, the same prompt retries on Groq and participants never notice.