Landing page
Feature summary
Doctor landing page
AI powered voice assistant
Appointment booking
Appointment schedule
Architecture Diagram

Inspiration

India's healthcare system faces a critical communication crisis. With over 22 official languages and countless dialects, doctor-patient communication is fragmented.

I witnessed this firsthand when my grandmother struggled to explain her symptoms to a doctor who didn't speak her native language.

Appointments that should take 15 minutes stretched to 45, with crucial details lost in translation.

Over 70% of India's population lives in rural areas where access to multilingual healthcare is virtually non-existent. Doctors spend more time on paperwork than with patients, leading to rushed consultations.

MedScribe AI was born from a simple belief: language should never be a barrier to quality healthcare.

What It Does

MedScribe AI is a voice-first healthcare communication platform designed for India’s multilingual reality.

For Patients

Speak Naturally
Converse in any of 8 Indian languages (English, Hindi, Tamil, Telugu, Bengali, Marathi, Gujarati, Kannada)
Health Assistant
Get instant answers about symptoms, medications, and medical reports
Voice Mode
Hands-free interaction for elderly users and people with limited literacy
Pre-Consultation History
Share medical history conversationally and save 10–15 minutes per consultation
Easy Scheduling
Book appointments using natural language

For Doctors

Smart Documentation
Auto-generate SOAP notes, prescriptions, and medical records
Pre-Consultation Insights
Review AI-collected patient histories before appointments
Voice Transcription
Real-time consultation recording with speaker separation
Appointment Dashboard
Streamlined scheduling and patient management
Clinical Codes
Automatic ICD-10 and CPT code suggestions

How We Built It

Frontend Architecture

Next.js 14 with App Router
TypeScript for type safety
Tailwind CSS with shadcn/ui
Framer Motion targeting 60 FPS

AI and Voice Technology

Real-time speech recognition using Web Speech API
Seamless switching between 8 languages
Groq API with Llama 3.3 (70B) for fast inference
Browser-native text-to-speech with fallbacks

Smart Features

Custom React hooks for voice state and transcription
Context-aware conversation history
Medical prompt tuning for clinical accuracy
Speaker diarization for consultations

State Management

Zustand for global state
React Context for preferences
Local storage for demo persistence

Challenges We Faced

1. Indian Accent Recognition

Initial accuracy was around 65%, which was unacceptable.

Solution:
A confidence threshold system was introduced. When certainty dropped below an acceptable level, the system asked for clarification. Accuracy improved to 85%.