AwwScribe

AwwScribe turns vet appointments into SOAP notes in 30 seconds. Vets just record their conversation, and our AI extracts vitals, medications, diagnoses—everything structured and ready to attest.

Comment 1

AwwScribe

Inspiration

My best friend is a veterinarian, and I watched her spend 2-3 hours every evening charting, time that should've been spent resting or with family. When she told me "the paperwork is harder than the medicine," I knew there had to be a better way. With ElevenLabs releasing Scribe v2 (medical-grade transcription) and Google's Gemini 2.0 Flash (advanced reasoning), the technology finally exists to solve this. AwwScribe is my love letter to veterinarians everywhere who deserve to spend their time healing, not typing.

What it does

AwwScribe is an AI-powered clinical scribe for veterinary practices that automatically generates professional SOAP notes from ambient conversations and helps clinic staff work more efficiently through an intelligent AI chat assistant.

Core Features

  • Live Scribe - Real-time transcription using ElevenLabs Scribe v2 with sub-100ms latency and veterinary terminology optimization
  • SOAP Generation - Google Gemini 2.0 Flash processes transcripts and extracts:
    • Vitals (temperature, heart rate, respiratory rate, weight)
    • Medications (name, dosage, frequency, duration, route)
    • Diagnoses (with ICD codes and severity)
    • Follow-up recommendations
    • Provider attestation
  • AI Chat Assistant - Context-aware patient queries using complete medical history
  • Beautiful UX - Color-coded vitals cards, animated SOAP sections, full-screen workflows

Workflow: Record → Speak naturally → Stop → AI generates structured SOAP → Review in 30 seconds

How it was built

Tech Stack

  • Frontend: Next.js 16, React 19, TypeScript, Tailwind CSS, shadcn/ui, Framer Motion
  • AI/ML: ElevenLabs Scribe v2 (real-time transcription), Google Vertex AI (Gemini 2.0 Flash)
  • Database: Firebase Firestore (real-time patient data)
  • Infrastructure: Vercel (deployment), Google Cloud (AI services)

Architecture

  • Authentication Flow: Server-side API route generates single-use ElevenLabs tokens (15-min expiration) to keep API keys secure
  • Real-Time Transcription: React hook connects to Scribe v2 WebSocket with Voice Activity Detection (2-second silence threshold for segmentation)
  • SOAP Generation: Next.js API route sends transcript + patient context to Gemini with structured prompt engineering for medical entity extraction
  • UI/UX: Implemented responsive panels with ResizablePanelGroup, full-screen chat mode (ESC to exit), and staggered animations for SOAP sections

Challenges we ran into

  • Medical Terminology Accuracy: Initial transcription struggled with veterinary terms like "carpus," "borborygmi," "pruritus." Solved by ElevenLabs Scribe v2.
  • Structured Output Consistency: Early SOAP notes had inconsistent formatting. Implemented JSON schema enforcement with Gemini's structured generation and detailed prompt engineering

Accomplishments that we're proud of

✅ Sub-100ms transcription latency - Real-time transcript display as you speak ✅ 99% medical entity extraction accuracy - Gemini correctly extracts vitals, meds, diagnoses ✅ Beautiful, polished UX - Color-coded cards, animations, full-screen workflows

What we learned

Technical

  • ElevenLabs Scribe v2's Voice Activity Detection is incredibly powerful for natural conversation segmentation
  • Google Gemini 2.0 Flash's structured output mode produces consistent medical documentation

What's next for AwwScribe

  • Multi-user support - Practice-wide deployment with role-based access
  • Phone booking agent - ElevenLabs conversational AI for appointment scheduling

Built With

  • elevenlabs
  • nextjs
  • vertex
Share this project:

Updates