Serenity

Serenity is your AI therapy companion, she remembers you, and offers 24/7 emotional support using voice, memory and proven mental health techniques

Comment

Inspiration

I've personally struggled with mental health going through times when I had no one to talk to. And I know I'm not alone. Every day, countless people silently suffer, often unable to express their feelings, and some never get the chance to.

Mental health support shouldn't be a luxury. It should be freely accessible, available anytime, and feel safe and human. That’s why I built Serenity, your AI therapy companion who remembers, understands, and cares.

What it does

Serenity is an AI-powered therapy companion designed to listen, understand, and provide emotional support, anytime, anywhere. Serenity can:

  • Hold meaningful conversations with empathy and warmth
  • Remember user stories and context using memory embeddings
  • Offer support using real therapeutic frameworks like CBT, ACT, and mindfulness
  • Talk to users via voice (powered by ElevenLabs) for a more personal connection
  • Provide personalized help, even when you can’t talk to a real person

Serenity is not just another chatbot. She's designed to be a safe space, a friend, and a gentle guide during hard times

How we built it

  • Frontend: Built with Next.js for a responsive, modern UI
  • Backend: Used Supabase for authentication, storage, and vector search
  • Memory System: Integrated a vector database in Supabase to store long-term memory and emotional context for each user
  • AI Core: Uses LLMs for contextual, emotionally intelligent responses with therapeutic alignment
  • Voice: Integrated ElevenLabs for realistic two-way voice conversations
  • Design & Persona: Created a warm, animated AI persona you can emotionally connect with

Challenges we ran into

  • Humanizing the AI: Making Serenity genuinely empathetic and comforting, not robotic, required tons of experimentation with prompts and emotional memory handling.
  • Vector memory complexity: Building a meaningful memory system with Supabase's vector support was challenging, tuning the embeddings and matching felt more like therapy than tech.
  • Latency in voice conversations: Real time voice support with ElevenLabs took fine tuning to maintain emotional flow and minimize response lag

Accomplishments that we're proud of

  • Serenity feels real. Users connect with her. She remembers, comforts, and supports.
  • Successfully built a working memory system with embeddings that simulate long-term recall and emotional continuity.
  • Created a voice-capable therapy AI that's gentle, non judgmental, and always available.
  • Onboarded tons of free users in a 2 day launch among friends!!!

What we learned

  • Emotional design matters just as much as technical architecture when building AI companions.
  • Vector databases can be used powerfully to simulate memory and improve contextual intelligence.
  • It's possible to make an AI feel warm and safe if you deeply care about the user's emotional journey.

What's next for Serenity

  • Launching full-duplex voice mode for truly natural conversation
  • Training deeper emotional models fine-tuned for mental health situations
  • Expanding memory capabilities with user consent for deeper personalization
  • Mobile app for on-the-go emotional check-ins
  • Partnering with therapists to offer real time escalation if users are in distress

Mental health should be free, accessible, and kind, and Serenity is one step closer to making that possible.

Built With

  • elevenlabs
  • gemini
  • nextjs
  • supabase
  • tailwind
Share this project:

Updates