🌱 Inspiration

Mental health support is often expensive, inaccessible, or intimidating. Many students and professionals struggle to find someone to talk to when they need it most.
We wanted to build a voice-first AI therapist that feels more natural than a chatbot, where users can simply speak and listen instead of typing. The goal was not to replace therapy, but to lower barriers to entry and provide a safe, private, and always-available companion.

🛠️ How We Built It

  • Frontend (React + Hooks): Implemented microphone recording, live captions, and audio playback with a clean, modern UI.
  • Backend (FastAPI + Python): Designed REST endpoints for TTS, STT, and Chat. Used a lightweight TTS model (Kokoro) and mock therapy responses for prototyping.
  • Voice Loop: Connected recording → STT → AI response → TTS → playback, creating a seamless cycle.
  • Session Context: Implemented message history to preserve conversational flow.

📚 What We Learned

  • How to integrate speech pipelines (STT + TTS) into real-time applications.
  • Importance of session flow design (handling context, retries, and cleanup).
  • Challenges of balancing latency, accuracy, and naturalness in voice-based AI.
  • Learned that sometimes a mocked flow is enough to demo a powerful concept without full production APIs.

🚧 Challenges

  • Latency: Getting audio responses fast enough to feel conversational.
  • STT Accuracy: Current transcription is mocked; Whisper/Google API integration is pending.
  • Scalability: Managing temporary audio files and cleanup under heavy use.
  • Privacy: Designing for HIPAA compliance and anonymization in the future.

🔮 Next Steps

  • Swap mock therapy responses with GPT-4 or other LLMs fine-tuned for CBT.
  • Replace STT mock with Whisper / Google Cloud Speech-to-Text.
  • Add avatar generation for a more engaging human-like presence.
  • Explore emotion detection and multi-language support.

This project showed us that building a realistic AI voice therapist is not only possible, but can be approachable in a hackathon setting by focusing on the core user experience loop first.

Built With

Share this project:

Updates