MindBridge AI

24/7 AI mental health companion with voice support, multi-agent therapy routing, and smart journaling - always there when you need someone to talk to.

Comment 2

Inspiration

Mental health crises don't wait for office hours. 1 in 5 adults experience mental illness yearly, yet 75% of mental health conditions go untreated. The 988 Suicide Hotline receives 4.8 million calls annually in the US alone, with wait times that can feel like an eternity during a crisis.

I built MindBridge AI because I believe everyone deserves access to compassionate support at 2 AM, during a panic attack, or whenever life feels overwhelming. Not as a replacement for professional help, but as a bridge - always available, never judgmental, ready to listen.

What it does

MindBridge AI is a multi-agent mental health support platform with:

🎙️ Voice Companion - Talk naturally using speech recognition, with responses spoken back using ElevenLabs for natural, empathetic voice synthesis. Different voices for different agents (calming for sleep support, grounded for crisis).

💬 Multi-Agent Chat - A triage system routes conversations to specialized AI agents:

  • CBT Agent - Cognitive behavioral therapy techniques
  • Sleep Agent - Insomnia and sleep hygiene support
  • Crisis Agent - Immediate assessment and de-escalation
  • Grounding Agent - Anxiety and panic grounding exercises

📝 Smart Journal - Write daily entries with AI-powered sentiment analysis (Cerebras) that tracks your emotional patterns over time.

📊 Insights Dashboard - View mood trends, session statistics, and query your data with natural language ("Show me entries where I felt happy").

📄 Document Upload - Store therapy resources with Vultr Object Storage, searchable via SmartBuckets RAG.

How we built it

Backend (Raindrop Platform):

  • 14 microservices deployed on LiquidMetal Raindrop
  • SmartMemory for conversation context across sessions
  • SmartBuckets for document RAG search
  • PRIMARY_DB (SmartSQL) for structured data storage
  • All services communicate via Raindrop service-to-service calls

Frontend (Next.js 15):

  • Deployed on Vercel
  • shadcn/ui components for accessible, beautiful UI
  • Web Speech API for voice input
  • Real-time state management

External Integrations:

  • Cerebras - Ultra-fast AI inference for sentiment analysis and response generation
  • ElevenLabs - Natural text-to-speech with emotional tone
  • Vultr Object Storage - S3-compatible document storage
  • Stripe - Payment processing (test mode)
  • WorkOS - Authentication infrastructure

Challenges we ran into

  1. Edge Runtime Constraints - Raindrop runs on Cloudflare Workers, which meant no TCP connections. Had to adapt all external API calls to HTTP-based services.

  2. Multi-Agent Coordination - Building a triage system that correctly routes users to the right specialist agent based on crisis level and topic detection required careful prompt engineering.

  3. Voice + AI Latency - Coordinating speech recognition → AI response → TTS synthesis without awkward pauses. Cerebras helped significantly with fast inference.

  4. SmartMemory Learning Curve - Understanding the working memory vs episodic memory paradigm took experimentation to implement correctly.

Accomplishments that we're proud of

  • 14 deployed services working together seamlessly on Raindrop
  • Real voice conversations with ElevenLabs - it actually feels like talking to someone
  • Multi-agent routing that intelligently escalates crisis situations
  • Natural language insights - asking "show my happiest journal entries" and getting real results
  • Built the entire platform solo in under a week

What we learned

  • Raindrop's SmartComponents (especially SmartMemory and SmartBuckets) dramatically simplify building AI applications
  • Edge computing constraints force better architecture decisions
  • Voice AI requires careful attention to latency at every step
  • Mental health AI needs extra care around safety - always provide crisis resources, never replace professional help

What's next for MindBridge AI

  • Real-time voice streaming - continuous conversation without push-to-talk
  • Therapist dashboard - allow mental health professionals to monitor patients (with consent)
  • Mobile app - native iOS/Android for always-available support
  • Crisis hotline integration - seamless handoff to human support when needed
  • Multilingual support - mental health support in every language

Built With

  • cerebras
  • elevenlabs
  • next.js
  • raindrop
  • shadcn-ui
  • smartbuckets
  • smartmemory
  • smartsql
  • stripe
  • tailwind-css
  • typescript
  • vercel
  • vultr-object-storage
  • workos
Share this project:

Updates

Private user

Private user posted an update —

Currently, the frontend uses a shared demo user ID (demo_user_001) across all sessions to ensure features can be tested with existing data. To enable user-specific data isolation, the frontend simply needs to use the authenticated user's ID from the login session. Backend already supports per-user data filtering. This is a minor frontend configuration change.

Log in or sign up for Devpost to join the conversation.

Private user

Private user posted an update —

During the original demo recording, ElevenLabs free tier quota was exceeded. No code changes were made - we simply upgraded to the ElevenLabs Starter plan, and the voice feature now works as intended.

Log in or sign up for Devpost to join the conversation.