RUTHRAN RAGHAVAN
Case Study|Healthcare / Mental Wellness

PsyPal — Voice AI Mental Health Companion

A real-time, voice-to-voice AI companion offering person-centered emotional support in 15+ languages.

The Problem

Understanding the business challenge.

Access to timely mental health support is limited by cost, availability, and stigma.

Text-based chatbots feel impersonal and lack the immediacy of spoken conversation.

Most support tools don't let users speak in the language they're most comfortable discussing sensitive topics in.

The Solution

Our AI-powered approach.

A real-time, voice-to-voice AI companion grounded in Carl Rogers' Person-Centered Therapy approach, enabling natural spoken conversations across 15+ languages with mood tracking and crisis escalation.

Real-Time Voice Conversations

Bidirectional audio streaming via the Gemini Live API for natural, low-latency spoken sessions.

15+ Language Support

Native conversation support across English (US/UK/India), Hindi, Tamil, Telugu, Kannada, Malayalam, Bengali, Gujarati, Marathi, Punjabi, Odia, Assamese, Urdu, Spanish, French, and German.

Mood Dashboard

Visual tracking of daily sentiment and anxiety trends over time using interactive charts.

Emergency SOS Routing

Routes users to crisis helplines when a conversation indicates acute risk.

Guided Exercises & Journaling

Structured breathing exercises and reflective journaling alongside conversational support.

Technical Architecture

Enterprise-grade technology stack.

Frontend
React 19 with Tailwind CSS v4 and Recharts for mood visualization
Backend
Node.js, Express, and WebSockets (ws) for real-time audio relay
Vector Database
Not applicable — real-time conversational AI, no document retrieval
Models
Google Gemini Live API (gemini-3.1-flash-live-preview) for real-time bidirectional voice
Retrieval
N/A — session-based conversational context, not retrieval-augmented

Security & Compliance

Privacy-first implementation.

  • Voice audio streamed directly to Google's Gemini Live API for processing
  • Session-based interaction state with no long-term identity linkage required
  • Dedicated emergency SOS pathway for acute risk detection

Implementation Workflow

Structured deployment process.

1

Voice Capture

Browser captures microphone audio and encodes it to 16-bit PCM.

2

Real-Time Streaming

Audio is streamed via WebSocket to the Gemini Live API for processing.

3

Empathetic Response Generation

The model generates a person-centered, spoken response.

4

Playback & Mood Logging

Synthesized voice (24kHz PCM) plays back while sentiment is logged to the mood dashboard.

Business Impact

Measurable outcomes.

Lowers the barrier to seeking emotional support by making it voice-first and available in 15+ languages
Provides continuous mood tracking that surfaces patterns users might not notice themselves
Enables a crisis escalation pathway that connects users to real help when needed
Removes the friction of typing during emotionally difficult conversations

Target Industries

Versatile application across sectors.

Healthcare
Mental Wellness
Telehealth

Experience it live

Try Live Demo