MediCore AI

Inspiration

Millions of people lack easy access to reliable, multilingual healthcare advice—especially during emergencies or when interpreting complex medical data. With the rise of LLMs and voice interfaces, we saw an opportunity to create a powerful yet accessible tool. MediCore AI is inspired by the need for a smart, secure, AI-driven assistant that empowers individuals and families to take control of their health, regardless of language or location.

What it does

MediCore AI is a comprehensive, privacy-first LLM-powered medical assistant that offers:

Authentication System

User registration & secure login
Demo accounts for testing
Route protection & session management
Logout with session termination

AI-Powered Medical Consultation

Google Gemini AI integration
Four languages: English, Hindi, Spanish, French
Multiple modes:
- General Medical Consultation
- Symptom Triage Assessment
- Emergency Medical Guidance
- Second Opinion Support
Real-time AI chat with context-aware responses

Advanced Voice Management

Voice input with auto language detection
Native TTS in all supported languages
Voice speed control & keyboard interruption
Smart voice mapping per language

Test Results Analysis

File upload (PDF, JPG, PNG, TXT up to 10MB)
AI-based value extraction and analysis
Reference range comparison and risk scoring
Personalized health recommendations
Manual entry and sample data demo

Emergency Services

GPS location detection
Nearby hospital locator
Emergency contact quick-dial
Emergency triage system with AI
Country-specific critical action protocols

Appointment Booking

Smart date & phone validations
Specialized booking by category
Time slot management & confirmation

Family Health Management

Multiple family profiles
Age-specific recommendations
Family health dashboard
Pediatric and elderly care guidance
Shared health history tracking

Health Records Management

Encrypted local file storage
Categorized document management
Search, filter, and export
Secure sharing with providers

Medication Management

Track doses and schedules
Reminders and refill alerts
Drug interaction checker
Side effect monitoring
Medication history logging

UI & UX Design

Dark/light theme toggle
Fully responsive design
Accessibility features (WCAG-compliant)
Modern UI with Tailwind & Radix UI
Smooth loading, error handling, offline support

Technical Highlights

Built with Next.js 14, React, TypeScript
Tailwind CSS & Radix UI components
Integrated Google Gemini AI
Progressive Web App (PWA) enabled
SEO-optimized, real-time updates

How we built it

Frontend: Next.js 14 App Router + React + Tailwind CSS + Radix UI
Voice Handling: Web Speech API for STT and TTS
LLM Integration: Google Gemini AI for conversational reasoning and medical analysis
Location & Emergency Tools: Geolocation API and hospital mapping with custom logic
Storage: Encrypted browser storage for sensitive data

Challenges we ran into

Real-time streaming of AI responses with voice playback
Multilingual voice synthesis with accurate language switching
Extracting structured data from unstructured lab reports
Designing a clean UX for multiple feature sets without overwhelming the user
Ensuring security without requiring backend databases

Accomplishments that we're proud of

Successfully built a multilingual AI health assistant powered by LLMs
Integrated real-time emergency triage and location-aware services
Created robust test result analysis from both file and manual input
Developed secure, encrypted, and offline-capable health record management
Delivered a smooth, responsive, and accessible UI across platforms

What we learned

LLM Fine-tuning for Medical Contexts: We learned how to guide general-purpose large language models like Google Gemini toward medical-specific reasoning using structured prompts and contextual memory.
Multilingual AI UX: Handling voice input/output across multiple languages taught us about pronunciation variances, language code detection, and mapping localized medical terminology to a common structure.
Voice + AI Integration: Combining streaming AI responses with real-time text-to-speech while allowing user interruptions (pause/stop/resume) required careful state management and event control.
Medical Data Extraction & Normalization: Parsing diverse test result formats (PDFs, images, raw text) helped us understand OCR challenges, medical unit conversions, and reference range normalization.
Privacy-First Design: Implementing a zero-backend, encrypted, local storage system taught us how to prioritize user privacy while still offering rich, persistent features like medical history, family data, and file uploads.
Human-Centered Health Design: Designing for all age groups—children, adults, and seniors—helped us focus on clarity, accessibility (WCAG), and emotional usability, especially during high-stress situations like emergencies.
Scalable UI Architecture: Managing a growing set of features (consultation, test analysis, booking, family management, emergency support) taught us modular design, route protection, and contextual rendering in Next.js App Router.
Progressive Web Apps (PWA): We explored how to deliver app-like experiences on the web using service workers, offline support, installability, and optimized load performance across mobile and desktop devices.
Error Handling in Conversational AI: We designed fallback flows and response validations to gracefully recover when the LLM misinterpreted symptoms, misunderstood context, or faced ambiguous user input.
Real-world Emergency Constraints: Simulating low-connectivity scenarios helped us design robust, fail-safe emergency support flows that fall back to static content or cached recommendations.
Healthcare UX Writing: Writing AI prompts and interface copy that were medically accurate and reassuring required a delicate balance of precision, tone, and empathy.
Data Validation for Medical Safety: Building input validations for names, dates, phone numbers, and report values made us appreciate the rigor required for safe, user-friendly health tech tools.