Practice difficult conversations with AI. Get real-time feedback on your body language and voice.

Built for Hack the Bay 2026 — USF Tampa

REPLAI puts you in high-stakes conversations — salary negotiations, job interviews, landlord disputes, HR complaints, legal client intakes. An AI opponent role-plays the other person while your webcam tracks eye contact, fidgeting, and posture, and your mic analyzes speech rate, filler words, and energy. After each round, you get a structured scorecard with specific feedback.

The Legal Client Intake scenario also runs real-time conflict-of-interest detection — extracting entity names from the conversation and fuzzy-matching them against a mock firm database.

git clone https://github.com/VickyAnaJ/HackTheBay.git
cd HackTheBay
npm install

Create .env.local:

NEXT_PUBLIC_GEMINI_API_KEY=your_gemini_key
LIVEKIT_URL=wss://your-livekit-url.livekit.cloud
LIVEKIT_API_KEY=your_livekit_key
LIVEKIT_API_SECRET=your_livekit_secret
NEXT_PUBLIC_LIVEKIT_URL=wss://your-livekit-url.livekit.cloud

Run:

npm run dev

Open Chrome at http://localhost:3000. Grant camera + mic access.

• Next.js 16 + React 19 + TypeScript + Tailwind CSS 4
• Google Gemini 2.5 Flash — conversation AI, coaching, entity extraction, structured feedback
• LiveKit WebRTC — mic input with echo cancellation
• MediaPipe Face Mesh — real-time body language tracking
• Three.js — 3D interactive background
• Custom conflict engine — Levenshtein + Jaccard fuzzy matching with risk scoring

• VickyAnaJ — Gemini integration, legal conflict system
• HarenG8374 — Body tracking, UI components, styling
• keillycespedes — LiveKit voice pipeline, session layout, 3D background
• 4shivv — Core types, TTS engine, page orchestrator, docs