AI-powered consultation tool for pro-bono advisors working with underserved communities across language and country barriers.

A consultant joins a WebRTC video call with their client. Risus handles everything else:

• Live translation — speech is transcribed, translated, and spoken back in the target language in near real-time (ElevenLabs Scribe → Gemini Flash → ElevenLabs TTS), across 99 languages
• AI projections — Gemini reads the transcript and generates only the sections relevant to that conversation: financial forecasts, risk radars, market charts, team capacity, community impact
• Country-aware analysis — the AI surfaces legal, banking, and social constraints specific to the client's country: wire transfer fees, currency exposure, local tax obligations, purchasing power, geopolitical risk
• PDF reports — download a formatted AI projection report; reports are saved locally and accessible from the Deliverables page
• Case repository — searchable, anonymised case studies with a points-based publishing incentive for consultants

Browser mic
  → MediaRecorder (audio/webm)
  → POST /api/translation/stt       (ElevenLabs Scribe)      → transcript
  → POST /api/translation/translate  (Gemini 2.5 Flash)      → translated text
  → POST /api/translation/tts       (ElevenLabs Turbo v2.5)  → audio blob
  → <audio> element plays

All API keys stay server-side in Next.js API routes — never exposed to the client. Per-step latency (STT / LLM / TTS ms) is displayed in the test UI so bottlenecks are immediately visible.

git clone <repo-url>
cd Risus
npm install

Create .env.local in the project root:

GEMINI_API_KEY=your_gemini_api_key
ELEVENLABS_API_KEY=your_elevenlabs_api_key

npm run dev

Open http://localhost:3000.

cd webrtc-testing/server
npm install
npm run dev

npm run dev      # Start dev server with Turbopack
npm run build    # Production build
npm run start    # Start production server
npm run lint     # ESLint (zero warnings policy)

The Analysis tab ships with five built-in transcripts that demo country-specific AI insights:

src/
├── app/
│   ├── (app)/
│   │   ├── session/           # Live call page (video + transcript + controls)
│   │   ├── analysis/          # AI projections
│   │   ├── deliverables/      # Case repository + saved PDF reports
│   │   └── translation-test/  # Translation pipeline test UI
│   └── api/
│       ├── projections/       # Gemini analysis endpoint
│       └── translation/
│           ├── stt/           # ElevenLabs Scribe
│           ├── translate/     # Gemini translation
│           └── tts/           # ElevenLabs TTS
├── components/
│   ├── calls/                 # Video tiles, controls, transcript panel
│   ├── workspace/             # AI projections, charts
│   ├── deliverables/          # Case cards, saved documents panel
│   └── ui/                    # Design system (Button, Card, MetricCard…)
└── lib/
    ├── types.ts
    ├── mock-data.ts
    └── utils.ts

webrtc-testing/
├── server/                    # Node.js + Socket.IO signalling server
└── client/                    # Vite + React WebRTC test client

• Streaming translation — WebSocket-based STT + TTS to reduce perceived latency below 500ms
• Persistent backend — database for session history, reports, and case records across devices
• In-call overlay — live translated captions and AI risk flags on the video screen during the call
• Bilingual document export — formatted Word/PDF deliverables for clients to take away
• Organisation dashboard — aggregate impact metrics across all sessions: clients served, languages used, outcomes tracked