1. Gemini (cérebro, ouvidos e voz num stream só)

- Criar a chave de API no Google AI Studio: https://aistudio.google.com/apikey
- Gemini Live API, visão geral: https://ai.google.dev/gemini-api/docs/live
- Guia da Live API, com áudio, VAD, interrupção e transcrição: https://ai.google.dev/gemini-api/docs/live-guide
- Ferramentas na Live API (function calling e Google Search): https://ai.google.dev/gemini-api/docs/live-tools
- Function calling: https://ai.google.dev/gemini-api/docs/function-calling
- Lista de vozes, onde está a Charon: https://ai.google.dev/gemini-api/docs/speech-generation
- Limites do plano gratuito, bom pra explicar a cota: https://ai.google.dev/gemini-api/docs/rate-limits
- SDK Python google-genai: https://github.com/googleapis/python-genai

2. MediaPipe (gestos pela webcam)

- Repositório: https://github.com/google-ai-edge/mediapipe
- Gesture Recognizer, com os 7 gestos prontos: https://ai.google.dev/edge/mediapipe/solutions/vision/gesture_recognizer
- Versão Web/JS, que é a usada no projeto: https://ai.google.dev/edge/mediapipe/solutions/vision/gesture_recognizer/web_js
- Hand Landmarker, os 21 pontos da mão: https://ai.google.dev/edge/mediapipe/solutions/vision/hand_landmarker
- Demo oficial no navegador, ótima pra mostrar na tela: https://mediapipe-studio.webapps.google.com/demo/gesture_recognizer

3. Áudio e câmera no navegador (APIs nativas, sem lib)

- getUserMedia: https://developer.mozilla.org/en-US/docs/Web/API/MediaDevices/getUserMedia
- AudioWorklet: https://developer.mozilla.org/en-US/docs/Web/API/AudioWorklet
- WebSocket: https://developer.mozilla.org/en-US/docs/Web/API/WebSocket

4. Backend e build

- FastAPI WebSockets: https://fastapi.tiangolo.com/advanced/websockets/
- Uvicorn: https://github.com/encode/uvicorn
- Vite: https://vite.dev/
- Fonte Orbitron do visual Stark: https://fonts.google.com/specimen/Orbitron