Resonance — Therapist-Led Pods, AI-Assisted

Structured group care. Private pod support. AI behind the scenes.

Resonance is a demo prototype for matching people into therapist-led support pods, keeping support going through private group chat, and giving therapists an AI copilot for notes, moderation, participation balance, and recap drafting.

Demo Flow

1. Landing (/) — Learn the product positioning: therapist-led pods with AI support.
2. Onboarding (/onboarding) — Record or type an intake reflection, then choose relevant topics and support preferences.
3. Pod Match (/match) — Gemini matches the member into a pod based on reflection context, not just broad categories.
4. Session Workspace (/session) — The therapist leads the pod session while the right panel acts as a therapist workspace for:
   • facilitator guidance recommendations
   • moderation and safety suggestions
   • participation balance signals
   • therapist review workflow cues
5. Summary + Journal (/summary) — Member-facing recap after therapist review, mocked progress history, and a private journal area.

Quick Start

npm install
npm run dev

Open http://localhost:3000. The app falls back to mock responses when API keys are missing.

API Keys (Optional)

Create a .env or .env.local file in the project root to enable real AI and voice services:

cp .env.example .env.local

Variable	Purpose
GEMINI_API_KEY	Gemini Flash for intake matching, therapist guidance suggestions, safety analysis, and recap drafting
DEEPGRAM_API_KEY	Deepgram Speech-to-Text for onboarding/session transcription and Text-to-Speech support

All features work without keys using mock responses that look real.

Architecture

src/
├── app/
│   ├── page.tsx                  # Landing page
│   ├── onboarding/page.tsx       # Voice onboarding
│   ├── match/page.tsx            # Pod match
│   ├── session/page.tsx          # Therapist-led pod session + AI workspace
│   ├── summary/page.tsx          # Therapist-reviewed member recap + private journal
│   └── api/
│       ├── transcribe/route.ts   # Deepgram Speech-to-Text (non-streaming)
│       ├── speak/route.ts        # Deepgram Text-to-Speech proxy
│       ├── match/route.ts        # Gemini pod matching
│       ├── facilitate/route.ts   # Gemini therapist guidance suggestions
│       ├── safety/route.ts       # Gemini safety analysis
│       └── summary/route.ts      # Gemini recap drafting
└── lib/
    ├── ai/
    │   ├── facilitator.ts        # generateFacilitatorPrompt()
    │   ├── safety.ts             # analyzeSafety()
    │   └── summary.ts            # generateSessionSummary()
    ├── voice/
    │   ├── onboardingTranscription.ts  # Non-real-time transcription
    │   ├── liveTranscription.ts        # Real-time/simulated transcription
    │   └── textToSpeech.ts             # Spoken facilitator (Deepgram TTS / Web Speech)
    └── demo/
        └── seedData.ts           # Demo pod, therapist, recap, and progress data

Tech Stack

• Next.js 16 + TypeScript + Tailwind CSS + shadcn/ui
• Gemini Flash — intake matching, therapist guidance suggestions, safety analysis, recap drafting
• Deepgram Speech-to-Text — async onboarding and session transcription
• Deepgram Text-to-Speech — available for voice support experiments in the demo

Design Principles

• No real auth, no real DB — this is a demo, not production
• Every AI call has a mock fallback — demo works completely offline
• Therapist remains in charge — AI recommends actions, but the therapist leads the session and reviews member-facing notes
• Voice-first intake — onboarding uses audio recording; sessions support mic input
• Private journaling — journal content is member-only in the demo