Inspiration
TheraSphere started from something deeply personal.
I went to therapy, but I couldn’t open up. Not fully. I’d sit in silence or default to surface-level thoughts, knowing there was more underneath, but not knowing how to express it.
I wasn’t alone. Friends of mine were going through the same thing.
We weren’t against therapy. We just needed a bridge, a way to connect our thoughts to words before facing another person.
That’s when I started building TheraSphere.
Not as a replacement for therapy, but as a companion to it. A place where AI isn’t clinical, but quietly supportive.
Now it’s more than just a tool for those who struggle to open up. It’s for anyone who wants to understand themselves better, whether they’re in therapy, preparing for it, or simply checking in with their own mind.
What it does
TheraSphere is a mental health companion designed to support anyone navigating their emotional life, whether you’re preparing for therapy, already in it, or simply trying to make sense of your thoughts and feelings.
How I built it
TheraSphere uses a full-stack architecture built for speed, privacy, and emotional safety:
- Frontend: React + TypeScript for a highly interactive and type-safe user experience.
- Backend: FastAPI running on Uvicorn, enabling async communication and low-latency LLM integration.
- LLM Integration: Uses multiple LLMs
- Security: End-to-end encryption (E2EE) is used in chat to provide the most secure transmission.
Challenges we ran into
- Coordinating multiple LLMs - some are better at summarizing, others at soft prompting; orchestrating them took time and experimentation.
- E2EE implementation - especially for AI chat, ensuring full encryption while maintaining model usability required extra architectural care.
Accomplishments that I'm proud of
- Built a fully functional and E2E encrypted AI chatbot
- Created a space where users say they feel safe expressing things they wouldn’t say out loud.
But most importantly: I turned the silence I once felt in therapy into a product that helps others speak, even when it’s hard.
What we learned
- Trust is everything in emotional tools. That means designing for safety, not just utility.
- E2EE for AI isn’t common, but it should be. Mental health data should never be exposed, even in inference pipelines.
What's next for TheraSphere
- Voice journaling with tone-aware transcription and reflection.
- Fine-tuned emotional reflection models for richer feedback that feels even more personal.
- Therapist Bridge Mode, letting users securely share selected insights with professionals.
Log in or sign up for Devpost to join the conversation.