Grilled 🔥

The prep partner that sees you, hears you, and knows the material.

Grilled is a real-time AI interview and exam prep agent powered by Gemini 3.1 Flash Live API. It conducts voice-driven mock interviews and oral exams using your camera and microphone — watching your body language, detecting hesitation, and adapting questions in real time. When the session ends, it generates a structured gap report with AI-generated visual study aids and a voice debrief.

• Two modes: Interview Prep (mock interviews) and Exam Buddy (oral exam practice)
• Natural language goal input: Describe what you're preparing for in plain English
• Real-time multimodal session: Camera + mic with bidirectional audio via Gemini Live API
• Adaptive behaviour: Interrupts when you're wrong, encourages when you hesitate, escalates when you're strong
• Agentic resource search: Upload PDFs or let Gemini find materials via web grounding
• Gap report: Three-tier breakdown (strong / needs review / priority) with readiness score
• Visual cram cards: AI-generated study aid diagrams for your weakest topics (Gemini image generation)

• Frontend: React + Vite + TypeScript + Tailwind CSS
• Core engine: Gemini Live API — real-time bidirectional audio + video frames via WebSocket
• Web grounding: Gemini built-in agentic search
• Visual study aids: Gemini image generation — study diagrams from gap report topics
• Gap report: Gemini structured output — three-tier analysis with readiness score
• PDF parsing: PDF.js (client-side)
• Media capture: Browser MediaDevices API

No backend. No database. No auth. Everything runs client-side.

git clone https://github.com/your-username/grilled.git
cd grilled
npm install

Create a .env file:

cp .env.example .env

Add your API key:

VITE_GEMINI_API_KEY=your_gemini_api_key

Run the dev server:

npm run dev

Open http://localhost:5173 in your browser. Grant camera and microphone permissions when prompted.

src/
├── App.tsx                     # Router: landing → goal → resources → session → report
├── components/
│   ├── ModeSelect.tsx          # Interview Prep / Exam Buddy toggle
│   ├── GoalInput.tsx           # Natural language goal input
│   ├── ResourceUpload.tsx      # PDF upload + auto-search option
│   ├── ResourceConfirm.tsx     # Gemini confirms context before session
│   ├── Session.tsx             # Live session (camera, waveform, timer, end button)
│   └── GapReport.tsx           # Three-tier report + Gemini-generated visuals
├── hooks/
│   ├── useMediaDevices.ts      # Camera + mic access, video frame capture
│   ├── useGeminiSession.ts     # Live API session lifecycle + audio playback
│   ├── useAudioStreamer.ts     # Mic audio → base64 PCM chunks
│   └── useGapReport.ts        # Post-session gap report generation
├── lib/
│   ├── gemini.ts              # Gemini Live API WebSocket + standard completion
│   ├── gemini.ts              # Gemini Live API + completions + image gen
│   ├── pdf.ts                 # PDF.js text extraction
│   └── prompts.ts             # System instructions for interview/exam personas
└── types/
    └── index.ts               # TypeScript types

1. Choose mode — Interview Prep or Exam Buddy
2. Describe your goal — "Google L4 Frontend interview tomorrow"
3. Add materials — Upload PDFs or let Gemini search for you
4. Live session — Gemini asks questions via voice while watching your camera feed
5. Gap report — Structured breakdown of strengths and weaknesses
6. Visual cram cards — Gemini generates study aid diagrams for weak topics

Built at {Tech: Europe} London AI Hackathon (April 2026) for the Google DeepMind — Real-Time Conversational Agents with Gemini 3 track.

• Google DeepMind — Gemini Live API (real-time multimodal), Gemini Flash (gap report), Gemini image generation (visual cram cards)

• Gemini API by Google DeepMind