Community Spotlight: Building Voice AI for Mental Health Support

This is the first in our Community Spotlight series, featuring developers building remarkable things with Ferni.

Meet Sarah Chen, a developer who left her role at a major tech company to build MindfulMoments - a voice-first therapy companion that's now helping over 10,000 users practice mindfulness and manage anxiety.

We sat down with Sarah to learn how she built it.

The Problem

"I was in therapy for years," Sarah explains. "My therapist was amazing, but I only saw her once a week. The other 167 hours? I was on my own."

Sarah noticed she'd forget the techniques between sessions. "I'd be in the middle of an anxiety spiral at 2 AM, and I couldn't remember what my therapist told me to do."

Why Voice?

"Typing during a panic attack is nearly impossible," Sarah says. "Your hands shake. You can't think clearly. But talking? Talking feels natural. Even at your worst, you can usually talk."

She tried building a chatbot first. "It helped, but something was missing. The interaction felt clinical, not supportive. Then I discovered Ferni."

The Technical Journey

Starting Simple

Sarah's first version was minimal:

import { FerniClient } from '@ferni/sdk';

const client = new FerniClient({
  apiKey: process.env.FERNI_API_KEY,
  persona: 'maya',  // Calm, supportive coaching persona
});

client.registerTool({
  name: 'guided_breathing',
  description: 'Start a breathing exercise',
  handler: async ({ duration }) => {
    return {
      type: 'breathing_exercise',
      pattern: '4-7-8',  // Inhale 4s, hold 7s, exhale 8s
      duration,
    };
  },
});

"That first version could do guided breathing and basic grounding exercises. Nothing fancy. But users loved it."

Adding Emotional Intelligence

The breakthrough came when Sarah started using Ferni's emotion detection:

client.on('emotion_detected', async (emotion, context) => {
  if (emotion.anxiety > 0.7) {
    // User is highly anxious - switch to grounding mode
    return {
      response: "I can hear you're feeling overwhelmed right now. " +
                "Let's pause together. Can you tell me five things you can see?",
      mode: 'grounding',
      technique: '5-4-3-2-1',
    };
  }
});

"The first time MindfulMoments detected my anxiety before I explicitly said anything, I cried. It felt like someone actually understood."

Context That Remembers

Users don't want to re-explain their situation every session:

// Store therapeutic context
client.registerTool({
  name: 'update_context',
  handler: async ({ key, value }, context) => {
    await context.memory.store({
      category: 'therapeutic_context',
      key,
      value,
      // Never auto-expire therapeutic insights
      ttl: null,
    });
  },
});

// Use context in responses
client.on('session_start', async (context) => {
  const recentTriggers = await context.memory.get('recent_triggers');
  const copingStrategies = await context.memory.get('effective_strategies');

  return {
    systemContext: `
      User's recent anxiety triggers: ${recentTriggers}
      Strategies that have worked for them: ${copingStrategies}
      Approach: Validate feelings, then gently suggest proven strategies
    `,
  };
});

Results

After 6 months:

"That 2-6 AM stat hit me hard," Sarah says. "These are people who can't sleep because of anxiety. And they're getting help at the moment they need it most."

Lessons Learned

1. Voice Changes Everything

"Users tell me things they'd never type. There's something about speaking that bypasses the inner critic. People open up in ways that surprise them."

2. Silence is Powerful

// Don't fill every pause
client.configure({
  silenceHandling: {
    // Wait 5 seconds before speaking during exercises
    exerciseMode: 5000,
    // 3 seconds for normal conversation
    conversationMode: 3000,
    // Longer pauses are okay during emotional moments
    emotionalMode: 8000,
  },
});

"In therapy, silence is healing. I had to teach MindfulMoments that not every pause needs to be filled."

3. Escalation Paths Matter

client.on('crisis_detected', async (context) => {
  // This is critical - know your limits
  return {
    response: "I'm really glad you shared that with me. " +
              "What you're going through sounds serious, and I think " +
              "it would help to talk to a crisis counselor. " +
              "Can I connect you with someone right now?",
    action: 'offer_crisis_resources',
    resources: [
      { name: '988 Suicide & Crisis Lifeline', phone: '988' },
      { name: 'Crisis Text Line', text: 'HOME to 741741' },
    ],
  };
});

"I'm not a replacement for therapy. I'm a complement. Knowing when to escalate is the most important feature I built."

What's Next

Sarah is working on:

1. Therapist Integration - Let licensed therapists review session summaries (with user consent)
2. Group Sessions - Guided group meditations with voice AI facilitation
3. Research Partnerships - Academic studies on voice AI for mental health

Try It

MindfulMoments is available at mindfulmoments.app.

Build Your Own

Inspired to build something similar? Here's how to start:

1. Ferni Quick Start
2. Emotion Detection API
3. Building Wellness Apps

Want to be featured in a future Community Spotlight? Let us know what you're building.