🎯 RoundZ - AI Interview Platform Summary

This is a fully autonomous AI interview platform that conducts voice-based technical screenings and auto-generates candidate scores without human involvement. The system uses Vapi.ai (GPT-3.5 + ElevenLabs voice + Deepgram transcription) to conduct 15-minute structured interviews across cultural fit, code review (SQL/Python), and logic puzzles. Transcripts auto-save to Supabase every 30 seconds, then a GPT-4 scoring agent analyzes them to produce objective scores (0-100) with detailed feedback. Built with React/TypeScript/Tailwind, it enables one person to scale from 50 to 2,000+ interviews/month at $0.20 per interview vs. traditional $60k/year recruiters. The demo shows natural voice conversations, live candidate video, on-screen code evaluation, and silent background processing. Key autonomous logic: interviews auto-start, transcripts upload on completion, scoring agent polls for pending reviews. Main failure modes: Vapi 400 errors (migrating to pre-configured assistants), network drops (localStorage backup implemented), API rate limits (queue system needed). Next steps: fix Vapi config, deploy scoring agent, build HR dashboard to review top 10% candidates.

Technologies: React, TypeScript, Vite, Tailwind, Vapi.ai SDK, Supabase, OpenAI GPT-3.5/GPT-4, ElevenLabs, Deepgram, WebRTC, MediaDevices API, PostgreSQL/JSONB, Qrant

We created everything today during the hackathon — the idea, the design, the entire codebase, the AI interview flow, and the full deployed product. There was no prior work, no existing code, and no prebuilt components before the start of the hackathon.