TheraBridge | Devpost

Inspiration

Rohin's story: I felt like my therapist was always starting from scratch with me. Every session began with "How are you doing?" and I'd struggle to remember what we talked about last week. I couldn't see my progress the way I could track fitness goals or work projects, even though it was happening. My coping techniques? Forgotten when I needed them at 2 AM. My breakthroughs? Lost in the silence between weekly appointments.

It got to the point where I couldn't see the point of therapy anymore. I wasn't motivated to show up. I almost became part of the 47% of patients who quit within the first three sessions - not because therapy doesn't work, but because the journey felt invisible.

We realized the problem wasn't just access to care - it was the silence between sessions. Therapy happens once a week for 50 minutes. But mental health challenges don't wait for appointments.

That's why we built TheraBridge.

What it does

TheraBridge transforms therapy sessions into structured progress and provides 24/7 AI support between appointments.

For Patients:

Automatic session analysis: Upload audio → AI transcribes, identifies speakers, extracts mood scores (0-10), main topics, coping techniques, and action items
Dobby - AI therapy companion: Available 24/7, knows your full therapy history, guides you through panic attacks, celebrates breakthroughs, detects crisis keywords and provides immediate resources (988 Lifeline)
Progress tracking: Mood trends over time, visual timeline with milestones, treatment goals with progress bars, learned techniques library

For Therapists (coming soon):

AI-generated session notes (saves 30-45 min/session)
Patient dashboard with risk flags
Outcome measurement and technique effectiveness tracking

How we built it

Frontend: Next.js 16 + React 19 + Tailwind CSS + Framer Motion Backend: FastAPI (Python) + OpenAI Whisper API + pyannote.audio 3.1 speaker diarization + GPT-5/GPT-5-mini Infrastructure: Supabase + Railway

Key innovations:

Pure AI reasoning (no hardcoded rules)
Context-aware companion (Dobby gets full therapy history injected—last 3 session transcripts, learned techniques, mood trends, treatment goals)
Crisis safety protocol (empathetic responses + 988 hotline + therapist notification)

Challenges we ran into

Audio transcription pipeline complexity → Integrating Whisper API, pyannote speaker diarization, and role detection into a reliable end-to-end pipeline took multiple iterations to handle edge cases (overlapping speech, audio quality variations, file format conversions)
Cascading AI outputs → Every dashboard element depends on AI-generated data that feeds into other AI systems. We built a dependency chain: transcription → mood analysis → topic extraction → progress metrics → Dobby context. Getting the JSON parsing, error handling, and system prompts working together was intricate coordination.
Patient-centered design → We obsessed over removing visual bloat. Every element had to answer: "Does this help therapy or distract from it?" We redesigned the dashboard 4 times to maximize therapeutic utility—progress visualization without overwhelming, insights without information overload.
Complementing therapy, not replacing it → We wanted perfect yin-yang balance. TherapyBridge works best when therapists use our session prep guidelines—Dobby surfaces patterns, therapists provide human insight. Getting that balance right required rethinking what AI should and shouldn't do in mental healthcare.

Accomplishments that we're proud of

Multi-level AI-generated UI - Not just text generation—the entire dashboard is AI-powered. System prompts generate mood scores → those scores populate graphs → graphs inform progress metrics → metrics feed into Dobby's context. Every visual element (mood trend charts, progress bars, timeline milestones) is dynamically generated from cascading AI outputs.
Patient-centered design that patients love - We obsessed over every pixel. The dashboard doesn't feel like a medical tool—it feels like a journey. Clean, minimal, emotionally resonant. This is what therapists wish they could show their patients.
Production-ready system in record time - Complete audio pipeline, 3 AI analysis systems, full-stack app, authentication.. This isn't a prototype—it's deployable.
Clinical accuracy that rivals human assessment - Crisis detection with empathetic safety protocols. Context-aware AI that remembers entire therapy journeys.

What we learned

Prompt engineering is an art (specificity drives consistency)
Everything needs heuristics on top of raw ML
UX matters more than features (users want to SEE progress)
Paper and pen can be the missing link in your workflow

What's next for TheraBridge

Immediate: Complete backend integration, user testing with real therapists, HIPAA compliance roadmap 3-6 months: Therapist dashboard, clinical validation study, beta launch (50 therapists) 6-12 months: Seed funding ($1-2M), scale to 500 therapists, EHR integrations 1-3 years: Enterprise partnerships, predictive analytics (suicide risk, dropout prediction), international expansion (Whisper supports 99 languages)

Dream feature: Real-time session insights whispered to therapists via earbud

Why this matters: 47% of patients quit therapy because progress feels invisible. TheraBridge makes healing visible. It gives patients a companion who remembers their journey and therapists tools to see patterns. It fills the silence between sessions with support, guidance, and hope.

We built this because healing shouldn't feel lonely.