SERENITY-AI-Therapist

Serenity AI reimagines therapy as a phone call, not an app. When a user taps “Call Serenity,” their iPhone rings
with a native CallKit incoming call — they answer it like any FaceTime, and drop into a full-screen video
session with an AI therapist powered by Tavus CVI. This isn’t a chat window with a webcam bolted on. The
interface is the innovation: the phone call is the entry point, the video session is the container, and Claude’s
reasoning drives everything happening beneath the surface in real time.

During the session, Claude Sonnet 4 listens to every therapist utterance via Tavus webhooks and generates
contextual micro-tips — categorized as coping, reflection, action, or grounding — that surface in a collapsible
overlay at the bottom of the video. These tips refresh every 15 seconds, transforming a passive video call into
a living, intelligence-driven experience where the UI itself responds to the emotional arc of the conversation.

After the session, Claude powers five interactive CopilotKit artifacts that shatter the standard chat-window
paradigm: a checkable Self-Care Plan with progress tracking, a multi-step 5-4-3-2-1 Grounding Exercise with
sensory prompts, an animated Breathing Exercise with expanding/contracting visual feedback, a 7-step CBT Thought
Record worksheet for cognitive reframing, and a structured Session Summary with topic chips and next steps.
These aren’t static text responses — they’re generative, interactive UI components that Claude populates with
deeply personal, context-aware content (“Set a boundary with your ex this week,” not “Practice self-care”).

Stability is achieved through pre-warmed conversations (Tavus sessions are created in the background before the
user ever taps “Call,” eliminating cold-start latency), graceful Redis degradation (the app functions fully
without a cache layer), and a Zustand-driven 5-phase state machine (home → ringing → connecting → inSession →
postSession) that ensures deterministic screen transitions. Session memory persists across calls — Claude
injects summaries of previous sessions into the therapist’s system prompt, so Serenity can say “Last time you
mentioned feeling isolated — how are you sitting with that now?” without any model fine-tuning.

The entire experience is tuned for a specific persona: a startup founder going through a breakup. The therapist
speaks in practical frameworks, validates emotions before offering action items, and uses founder-friendly
language. Claude doesn’t just generate content — it drives the generative experience end-to-end, from real-time
tip generation to interactive artifact creation to cross-session memory continuity.

• React Native + Expo — Cross-platform mobile app with native iOS feel
• react-native-callkeep — Native CallKit integration for real incoming/outgoing call simulation
• Zustand — Lightweight state management for 5-phase session orchestration
• Tavus CVI — Real-time conversational video AI (~600ms latency) with custom therapist persona
• Anthropic Claude Sonnet 4 — Real-time therapy tip generation, interactive artifact content, and chat reasoning
• CopilotKit — Structured LLM actions that render interactive React artifacts (self-care plans, grounding
  exercises, breathing animations, CBT worksheets, session summaries)
• Next.js 16 — Backend API routes, CopilotKit runtime, and webhook processing
• Redis (Upstash) — Session caching (25-min TTL), tip storage, memory persistence, artifact storage (30-day TTL)
• Google Cloud Run — Production backend deployment
• pino — Structured logging across all backend services
• WebView — Embeds Tavus video session within React Native with injected JavaScript for seamless full-screen experience

N/A all built today